Intelligence agencies for Australia, Canada, New Zealand, the United Kingdom and the United States warned in a rare joint statement that AI-powered cyberattacks capable of overwhelming government and business defenses may be just months away. 

The June 22 statement by the agencies—known as the Five Eyes—says they anticipate the advanced capabilities of frontier models like Anthropic’s Fable 5 and OpenAI’s Daybreak will “exceed current industry expectations, fundamentally transforming both offensive and defensive cyber capabilities.” 

The agencies did not publicly detail the evidence underlying their assessment, but the warning aligns with concerns public cybersecurity and AI experts have been raising for months. The statement comes just days after the U.S. government issued an export control directive to Anthropic to suspend all foreign nationals’ access to Fable 5 and Mythos 5, whether inside or outside the country. The order is one of the most wide-ranging government responses to the capabilities of an AI model to date. 

“We now estimate a narrow three-to-five month window for organizations to outpace the adversary before AI-driven exploits start to become the new norm,” warned Lee Klarich, chief technology officer of the cybersecurity company Palo Alto Networks in May. “This impending vulnerability deluge demands urgency.” 

The alliance’s warning targeted leaders not only of governments and corporations, but of small and medium businesses around the world, urging them to gauge their organizations’ risk levels, review readiness measures, and remain actively engaged with emerging AI-related threats. The agencies identified outdated systems, slow patch management, unnecessary internet connectivity, weak access controls and inadequate incident-response planning by organizations as vulnerabilities leaving organizations especially vulnerable to AI-enabled cyberattacks. 

The statement included a set of practical actions that leaders can use to strengthen their defenses, including limiting who and what can connect to systems, quickly installing security updates and replacing outdated and unsupported technology. The statement also advised strictly controlling who has access to sensitive or confidential information and to regularly practice response drills in case of an attack.

“Success will not come from having the most tools,” the intelligence agencies said. “It will come from getting the basics right, acting quickly, and integrating cyber security into core business strategy.”

At its core, the guidance reinforces a long-standing recommendation: cybersecurity should be treated as central to operational continuity rather than as a secondary concern. And, coming as it does so soon after the order to Anthropic, it reflects how rapidly threats are evolving as frontier AI systems advance and governments and companies struggle to keep step. 

“The rapid pace of frontier AI development means cyber risk assumptions can become outdated in months, not years,” the agencies said. “In that spirit, we call on leaders across industry to act now and work together to protect our people and secure our future.”

The post Western intel alliance warns ‘overwhelming’ cyberattacks could be months away  appeared first on The Media Copilot.

The European Commission has published a voluntary Code of Practice on Transparency of AI-Generated Content, giving AI providers and deployers a concrete path to compliance with the AI Act’s labeling requirements—and a clear reason to sign up.

The code, released June 10, 2026, covers two broad categories of obligations. Section 1 targets providers of generative AI systems, requiring them to mark outputs—audio, image, video, and text—in machine-readable formats and ensure their detection as artificially generated or manipulated. The technical solutions must be effective, interoperable, and reliable “as far as technically feasible,” factoring in content type, implementation costs, and the state of the art. Section 2 targets deployers, requiring them to label deepfakes (audio, image, or video that falsely appears authentic) and disclose AI-generated or manipulated text publications on matters of public interest.

The Commission also released a set of standard icons that deployers can use to label AI-generated content. Nicholas Diakopoulos, a professor at Northwestern University, shared them on LinkedIn:

The code is currently under adequacy assessment by the Commission and the AI Board. Once it clears that review, signatories can rely on its measures to demonstrate compliance with Article 50 of the AI Act, reducing administrative burden and gaining legal predictability across all EU member states. Non-signatories will have to demonstrate adequate compliance individually, assessed case-by-case by national market surveillance authorities.

Signatories also gain access to Signatory Taskforces: working groups set up to share implementation practices and advance marking and detection techniques across the value chain.

The code is described as a “consistent, practical and proportionate” implementation framework, not a replacement for the AI Act or the Commission’s forthcoming guidelines on Article 50’s scope.

The code was developed over three drafting rounds between September 2025 and June 2026, led by an independent chair and vice-chair. Participants included AI system providers, detection developers, industry associations, civil society organizations, academic experts, and organizations with expertise in transparency and very large online platforms. International and European observers also contributed without voting rights. Two dedicated working groups handled the providers and deployers tracks separately.

Key milestones included a first drafting round starting November 5, 2025, a second round in January 2026, a third round in March 2026, and a closing plenary on June 10, 2026—the same day the code was published.

The post EU publishes voluntary code on AI content transparency appeared first on The Media Copilot.

Look, I’m going to be straight with you. The traditional article is powerful. But it’s only one way to present your reporting.

You spend weeks on an investigation. You publish 3,000 sharp words. But what happens to the data behind it? The full timeline? The quotes that didn’t fit?

What if you could turn that investigation into an interactive experience, complete with clickable timelines, hover-activated charts and tagged insights, in about 20 minutes?

With vibe coding, it’s possible.

What is vibe coding?

The term vibe coding came out of developer culture, but it is no longer just for developers. It’s for anyone who wants to tell a story that harnesses the power of coding.

When you vibe code, you’re building an application with the help of AI by focusing on what you want it to do. Rather than coding with HTML, JavaScript or other technical languages, a builder describes the user experience in plain language to a Large Language Model (or LLM).

You might type a prompt like, “Build me an interactive timeline showing [x, y, z] events.”

Here’s what worked for the journalists I’ve seen succeed with vibe coding:

• Step 1: Pick something simple. Don’t try to rebuild your entire investigation. Start with one article, dataset or interview.
• Step 2: Use a basic prompt structure, like: “Build an interactive [website/dashboard/story] that shows [your content] in [style you want]. Focus on [what matters most].”
• Step 3: After you have something simple, iterate 3-5 times. First pass: structure. Second: visual style. Third: functionality. Fourth: polish.
• Step 4: Share your creation with a colleague. Don’t talk, just watch. See if they click around. If they get stuck, that means you built it for yourself, not your audience. Time to iterate again or start over.

Why vibe coding and journalism make sense

When I taught vibe coding through the Google News Initiative AI Lab, I watched journalists with zero coding experience build interactive financial dashboards, data visualizations and branded microsites, all in about 90 minutes.

“This would have taken our dev team a month,” one person told me. “I did it during our session.”

While you can move quickly to an initial application with vibe coding, you still want to get your product or development support staff on board before launching. The real benefit is that vibe coding lets you prototype faster to see if your idea works before needing to commit resources.

This matters because most newsrooms don’t have a developer on speed dial. At the Adirondack Explorer, a small regional outlet covering New York’s Adirondack Park, journalists are building a civic information product that aggregates town meeting recordings, transcripts and minutes across dozens of municipalities. That kind of project would normally require hiring contractors or a dedicated dev team. Instead, their reporters are building it themselves.

When I worked with VTDigger through the Google News Initiative, they automated campaign emails across four audience segments, work that directly generated $40,000 in donations that likely wouldn’t have happened with manual effort alone. Vibe coding turns “we can’t afford to build that” into “let me show you what I made this morning.”

Think of vibe coding as a creative prototyping partner. Get to 80% quickly. Then decide if you need developer support to get to 100%.

Tips for those who are new to vibe coding

I’m repeating this because it’s important: Start simple. Pick one piece of your big investigation. It could be a dense PDF that needs to be more accessible. It could be data sitting in a spreadsheet. It could even be an interview transcript with great quotes that couldn’t all make it into the article.

When describing what I want to the LLM, I get experimental with my prompts. I’ll type things like, “use the most modern UI and UX interactions and animations to make my charts and graphs more interesting and allow me to parse through the data visually.” Or, “build this in the style of a high-end investigative journalism piece meets Wired magazine’s data viz.”

Then I iterate with edits like, “change the color scheme to match our brand,” or, “pull out more quotes from the sources,” or even, “make [x, y, z] section more prominent.”

Three high-level no-no’s when vibe coding

1. Never use it for final production without verification.
2. Don't use it for anything requiring real-time data without a proper backend.
3. Never publish AI-generated content without independent verification. In high-stakes areas like health, legal, financial or public safety, errors can cause real harm.

A Google AI Overview recently told pancreatic cancer patients to avoid high-fat foods, which is the exact opposite of what oncologists recommend and could jeopardize a patient’s ability to tolerate chemotherapy.

AI can generate beautiful visualizations, but it can also confidently present wrong numbers. For anything where errors could harm your readers, verify everything against primary sources.

Vibe coding tools to try

The main platform I use for vibe coding is Lovable.dev. For a simple interactive graphic such as a timeline, searchable transcript or basic data visualization, you can expect to use roughly 3-8 credits to produce a solid prototype.

More complex builds with multiple views, filtering or light database features can take 15-30 credits depending on how much you iterate. The free tier is typically enough to experiment with small projects, while paid plans make sense if you’re building regularly or refining more advanced applications.

Bolt.new is another tool worth knowing. For a simple interactive project, such as a timeline or basic data visualization, you might use roughly 20,000 to 60,000 tokens depending on how much you iterate. More complex builds with custom logic, multiple components or repeated revisions can exceed 100,000 tokens. The free tier is generally sufficient for small experiments, while larger or ongoing projects may require a paid plan.

Bolt tends to give you more control over the code and works well if you want to edit things directly. Lovable is more beginner-friendly with a cleaner interface for non-technical users.

Both tools let you attach content like article text, CSV files or transcripts, describe what you want in plain language, and get a working prototype you can publish immediately.

You might wonder why you need these tools when you already have ChatGPT or Claude. The difference is output.

When you ask ChatGPT to build you a dashboard, it gives you code snippets you’d need to assemble yourself, often requiring a developer to make sense of it. When you ask Lovable the same thing, you get a working application with a live preview, hosting, and a chat interface to iterate on it.

Lovable is actually powered by Claude, but it wraps the AI in a full-stack builder that handles deployment, databases and design. For journalists without coding experience, that’s the difference between “here’s some code” and “here’s how it looks.”

Go build something new

The article format has served us well. It’s not dead, but it is not the only option.

When thousands of people fly to a conference, share incredible insights and then go home, that knowledge evaporates unless it’s transformed into something people can continue to engage with.

Vibe coding lets us do that better than text-only articles can. Not just for conferences, but for city council meetings, investigative data, community journalism and breaking news.

My hope is you’ll take these vibe coding tips and run. You’ll build interactive story formats your newsroom has never seen. You’ll prototype tools that solve real problems. You’ll make journalism more engaging, more accessible and more honest about its data and sources.

Journalists who can write, build, prototype, ship and transform their own work into new formats will define what news looks like in five years. So, go vibe code something. I can’t wait to see what you build.

This post first appeared in News Media Help Desk.

The post Vibe coding for journalists: Build interactive stories without writing a single line of code appeared first on The Media Copilot.

The Centre Daily Times in State College, PA, has voted to unionize after months of pushback against its parent company’s AI tool—a move that, according to The NewsGuild-CWA, makes it the first newsroom in the union to cite AI adoption concerns as a primary reason for organizing.

As Nieman Lab reported, the Centre Daily Times staff voted to join The NewsGuild of Greater Philadelphia last month. All eligible editorial staff signed authorization cards, and McClatchy voluntarily recognized the union. The catalyst, reporters told Nieman Lab, was McClatchy’s Content Scaling Agent (CSA) tool—an AI system that repackages existing articles into short-form summaries for publication or video scripts—and a March internal meeting where Kathy Vetter, McClatchy’s chief of staff for local news, told staff the company would use their bylines on AI-generated content unless union contracts prohibited it.

Josh Moyer, a senior reporter at the Centre Daily Times, took that as a signal. “It was essentially like, if you’re not in a union, your byline gets used; if you are in a union, we’ll follow what the union says,” Moyer told Nieman Lab. “If we want to control what happens to our byline, that’s the company telling us that we need to form a union.”

McClatchy introduced the CSA tool at the paper earlier this year. Reporters initially published at least one CSA-assisted story per week under a generic byline noting AI assistance. But in late February, management changed the policy: AI-generated content would now carry the reporter’s actual name. Reporters objected that it misrepresented their work to readers.

“When our names go on a thing, it says that this article or video is from that person, but that is just not true in this case,” said Trebor Maitin, a service reporter. Maitin was the first reporter at the paper to have his byline changed to reflect AI assistance.

The NewsGuild-CWA’s president, Jon Schleuss, said unionized newsrooms have had more success keeping AI content clearly labeled: “Unionized newsrooms are the ones where McClatchy’s AI slop gets a clear label. In non-union newsrooms, the AI slop may be carrying a real human reporter’s byline.”

Multiple McClatchy publications have seen byline strikes over the CSA tool, and some have taken labor actions over the tool and related workplace issues. For the Centre Daily Times, the union opens the door to formal collective bargaining and the ability to join coordinated actions at sister publications.

“Some of us use AI a lot more and are okay with it,” Maitin said. “But there is an overall understanding that we need to be able to have a say in this, and that unionizing at least gives us a seat at the table.”

The post A newspaper unionized because McClatchy put reporters’ names on AI content appeared first on The Media Copilot.

Reuters and Time are blocking all AI bots by default and only letting approved crawlers through—a whitelist approach that more publishers are adopting as the volume of unauthorized scraping grows.

As Digiday reports, both publishers moved to block AI bots last month, joining People Inc. and The Atlantic, which adopted similar strategies earlier this year and late last year respectively. The goal is simple: content costs money to produce, and AI companies have been taking it without paying.

“We saw that there was an imbalance between the value that publishers like Reuters provide and the value that Reuters receives in kind, and so instead we went from a default allow-all to a default disallow all,” said Josh London, head of Reuters Professional, which oversees the direct-to-consumer and direct-to-professional businesses. Reuters has since signed AI licensing agreements with Microsoft and Meta, according to the report.

The publishers aren’t relying on any single tool. Reuters uses robots.txt files, a method that is voluntary and non-binding, and one that many AI bots simply ignore. The approach is meant to create friction and signal that access requires negotiation. “If you want this, let’s have a conversation and then we can allow you to access,” said Alphonse Hardel, head of agency at Reuters, who leads the content licensing business.

Time allows roughly 70 bots on its site, ranging from AI lab crawlers and social platforms to its own operational systems. The volume of bot traffic has become significant enough that Time sees it as leverage for a future AI visibility product it’s developing for brand clients.

The economics are also shifting. Blocking bots cuts server costs: Hardel said the expense of the bot-blocking vendor can be nearly offset by the reduction in non-human traffic. At People Inc., the shift from a block list to an allow list meant going from blocking roughly 2,100 user agents to over 30,000, said Lindsay Van Kirk, the company’s SVP of innovation, speaking at an IAB Tech Lab event in May.

“Adding two full seconds of latency to the majority of scrapers when you implement a block-all-bots approach is a really good thing, even if they have to go through,” Van Kirk said. “Every scraper who has to pay a home proxy network in order to get access to the content is margin that you are taking out of their business.”

The IAB Tech Lab has published guidance on bot management, and the SPUR Coalition—a publisher group formed earlier this year with major news organizations—announced significant new membership as it works to create technical standards for AI licensing and content protection.

For Reuters, the change hasn’t reduced site traffic. After monitoring bot activity over an extended period, the company had enough data to identify which bots it could block without hurting revenue. The publisher maintains a public robots.txt file that lists approved bots, a benchmark that also supports enforcement discussions, said Phil Andraos, general manager of Reuters Digital.

“It’s not a set it and forget it approach,” London said. “The value of content is something that we ignore at our own peril, especially as AI scales.”

The post Reuters and Time flip the script on AI bots with blocking whitelists appeared first on The Media Copilot.

A German court has ruled that Google is directly liable for what its AI-generated search overviews say, in a decision that legal observers say could ripple far beyond Germany. As The Decoder reported, the Regional Court of Munich hit Google with a temporary injunction barring it from spreading false claims about two Munich-based publishers through its AI Overviews.

At the center of the ruling is a distinction the court drew sharply: AI Overviews are not search results. They are Google’s own content.

According to the court, Google’s AI Overviews had falsely tied the two publishing companies to scams, subscription traps, and shady business practices for certain search queries. The AI mixed up information about genuinely sketchy companies with the plaintiffs and drew connections that appeared in none of the linked sources. The publishers sent a cease-and-desist letter; Google didn’t respond appropriately, the court found.

The judges classified Google as a direct infringer because the overview “rewrites and judges results in its own words and according to its own structure.” In the case at hand, the AI opened with confident assertions like “Yes, [company] is known for dubious business practices,” then assembled its own summary, red flags, and user tips. Because Google built the AI, offered it, and controls its algorithms, the court ruled, Google owns what it produces.

Crucially, the court found that existing case law shielding search engines doesn’t apply. Germany’s Federal Court of Justice had previously granted traditional search engines limited liability because they merely point to outside websites. But AI Overviews generate “independent, new, and substantive statements,” the Munich court said, and only Google is positioned to check them against the underlying sources.

Google’s defense—that users can check the linked sources themselves and generally know not to blindly trust AI—fell flat. The court ruled that the ability to disprove a statement through further research doesn’t exempt a publisher from liability, drawing a parallel to press law, where outlets are liable for standalone teasers even if readers never click through. The reasoning is bolstered by research showing users almost never click source links in AI Overviews.

The court also weakened free speech protections for AI output, writing that an AI’s opinion is “not the expression of an acquired conviction” but “the result of an algorithm” and largely an expression of Google’s business interests.

Google was ordered to cover 80% of the legal costs, with the plaintiffs paying 10 percent each. The court said the ruling may have international reach.

The decision lands as scrutiny of AI accuracy intensifies. An analysis by AI startup Oumi for The New York Times found Google’s AI Overviews, running the current Gemini 3 model, answered correctly 91% of the time. At Google’s scale, that still means millions of wrong answers every hour—and a legal exposure that could extend to rivals like ChatGPT, Claude, and Perplexity.

The post German court rules Google is liable for false answers in AI Overviews appeared first on The Media Copilot.

In today’s competitive economy, nearly every organization aspires to be “data-driven,” but turning that ambition into measurable business outcomes remains inconsistent. Companies widely recognize the value of data and artificial intelligence, yet many are still struggling to operationalize these capabilities at scale.

When data foundations are weak, the effects extend beyond internal operations. Fragmented or unreliable data makes timely, well-informed decisions harder to reach, and increases the likelihood of gaps in areas like security and compliance. Ultimately, those gaps don’t stay internal. They affect the quality and consistency of customer experiences, and the confidence organizations can have in how they’re managing and protecting data responsibly.

Based on new global research from Cloudera, including a study conducted with Harvard Business Review Analytic Services, the gap is stark. Only 7% of enterprises say their data is fully ready for AI, while 27% report their data is not very ready or not at all ready. At the same time, expectations for transformation continue to accelerate, with organizations planning to embed AI across core business functions.

While companies are preparing for large-scale AI-driven transformation, most lack the underlying data infrastructure and maturity required to support it. Until that foundation is in place, the promise of AI remains difficult to fully realize.

Why Data Readiness Is So Difficult

Despite growing investment, data readiness has plateaued.

Enterprise data exists, but it can be hard to find or access because it is fragmented across systems. Over a third (34%) of respondents from the Data Readiness Index survey reported that siloed data was a major issue that prevented them from working together effectively to share, manage, and use data. These silos often stay in place because data isn’t well integrated across systems.

Most respondents said their data sources were somewhat integrated across various environments, but significant gaps remain. Only 30% of IT leaders reported that their data sources were fully integrated, while 52% said they were mostly integrated. While this shows some progress, it also highlights that many organizations are not yet fully prepared to support large-scale AI projects.   

Other barriers compound the problem. IT leaders also identified complicated access (47%), limited data visibility (44%), lack of training (41%), and cultural resistance (34%) as key obstacles.  Each issue slows progress, and together, they create systemic drag. At the same time, regulatory and security pressures are increasing. Data privacy and sovereignty requirements demand tighter control over how and where data is managed. In fragmented environments, meeting those requirements becomes more resource-intensive and more risky.

What “Data Readiness” Means In Practice

Data readiness ultimately comes down to trust and control. Organizations need confidence that their data is accurate, accessible, secure, and governed consistently, regardless of where it resides.

Governance is central to this goal. Findings from the Taming the Complexity of AI Data Readiness survey report show that organizations rank protecting sensitive data and privacy (59%), data quality (46%), and data governance (41%) as the most critical components of their data strategies. These priorities reflect a growing recognition that without strong governance, data cannot be trusted or effectively scaled across the enterprise.

At the same time, the Data Readiness Index reveals persistent structural challenges. Nearly a quarter of organizations (24%) report they cannot access all of their data across environments at any time, and 16% lack complete visibility into where their data resides. These gaps undermine governance at scale, making consistent policy enforcement unreliable and weakening an organization’s ability to manage risk.

Without trust and control, data can’t deliver value. Poor readiness delays insights and decisions as teams struggle to find and trust data. Disconnected environments harm customer experiences by blocking a unified view. Low-quality or poorly governed data leads to missed opportunities and higher risks.

When data is governed and secure, teams move faster and confidently, reducing validation time and increasing value. In the end, organizations must either operationalize data as a strategic asset or absorb the cost of its dysfunction.

A Widening Gap, And A Clear Opportunity

Data readiness is crucial for unlocking AI’s full potential, but readiness goes beyond simply collecting large amounts of data. Organizations also need systems that make trustworthy data connected and usable across the business. That includes improving data quality, establishing clearer governance and access controls, and creating visibility into where data comes from and how it moves through different systems.

These foundational efforts may happen behind the scenes, but they ultimately shape how effectively organizations can apply AI in the real world. In practice, the companies most likely to pull ahead may not be the ones adopting AI the fastest, but the ones building systems capable of delivering reliable, scalable outcomes over time.

This story was produced by Cloudera and reviewed and distributed by Stacker.

The post AI ambition rises as data readiness falls behind appeared first on The Media Copilot.

New research from more than 2,500 enterprise leaders finds the chatbot handling your support request has a better-than-even chance of having already been taken offline and restarted.

The AI-powered chatbot failed. The customer repeated themselves three times, got a confidently wrong answer, and gave up. For the company on the other end, that interaction didn’t just cost a support ticket but something harder to win back.

That scenario is playing out at scale. A new survey of 2,527 enterprise decision-makers across 10 countries conducted on May 12 finds that 74% of companies that deployed AI agents in customer communications have been forced to shut them down or roll them back, often after customers had already experienced the failure firsthand.

The research, published in May by communications infrastructure platform Sinch, examines a specific and underreported problem in the AI market: not whether companies can deploy AI in customer communications, but what happens after they do.

AI customer service has gone mainstream

There’s a widely accepted story in enterprise AI that the biggest challenge is getting pilots into production. McKinsey reported in 2025 that two-thirds of organizations remained stuck in experimentation phases. BCG found that 60% had yet to show any material value from their AI investments. Gartner forecast that half of all generative AI projects would be abandoned after proof of concept.

In customer communications specifically, something different happened. A Sinch study shows 62% of organizations already have AI agents live in production across customer channels, and 88% expect to be there by the end of 2026. That’s nearly 9 in 10 businesses actively deploying AI agents by the end of this year.

Enterprises aren’t dipping a toe in either. The average deployment spans 3.3 channels simultaneously, with nearly half running AI across four or more, including web chatbots, email, social media, WhatsApp, SMS/MMS, RCS, and voice and interactive voice responses. And the goal driving most of that investment isn’t cost reduction. For 36% of respondents, the primary objective is improving customer satisfaction and loyalty. They’re using AI to compete on customer experience, and to earn something harder to measure than efficiency: customer trust.

By every metric the market established to measure AI progress, these organizations won. They escaped pilot purgatory. They crossed the finish line.

Except that wasn’t the finish line.

Going live turned out to be the easy part

Here’s the finding that should make every AI communications leader stop and read more carefully: Research by Sinch from 2026 shows that 74% of organizations have been forced to shut down or roll back a live AI customer communications agent.

That holds across every region and every industry in the study. It doesn’t decline with experience. It doesn’t decline with investment. All along, the market has been drawing the wrong finish line, and what happens after enterprises successfully ship radically changes the question every AI communications leader should be asking right now.

More oversight hasn’t stopped the shutdowns

Here’s where it gets interesting. Among organizations that describe their guardrails as fully mature, the most governed, most monitored AI programs in the survey, the rollback rate is 81%.

More governance, more monitoring, more investment, and still 8 in 10 of the most advanced programs have had to shut something down.

The data offers a worrying explanation. Organizations with mature governance instrumentation can see failures that less mature organizations miss entirely. The programs reporting lower rollback rates aren’t necessarily running cleaner AI, and in many cases, they simply lack the monitoring to know when something goes wrong. The organizations reporting no governance failures are not the benchmark. They may just be the ones with the least visibility into what’s happening.

And then there’s the confidence problem: 90% of enterprise decision-makers describe themselves as confident in their AI agent readiness. Of those already in production, 75% have experienced at least one governance rollback. Confidence doesn’t correlate with fewer failures. In many cases, it’s the precise condition under which the next failure is being prepared.

The more useful question for any leadership team is, “If something went wrong right now, would we know before our customers did?”

When the chatbot goes down, brands feel it in three ways

When an AI communications agent fails in production, customers notice. The research shows the impact splits in three directions simultaneously, and most organizations are only tracking the first.

Research by Sinch from 2026 shows an increase in the support queue (35%) and reputational damage to the brand (34%) are the biggest impacts of AI agent failure.

Why support wait time spikes

Thirty-five percent of organizations cite a surge in human support agent load as the primary consequence. The agent goes down, and every interaction it was handling reverts to a human. A support team sized for a world where AI handles significant volume is suddenly managing all of it. At peak moments, a product launch, a service outage, a seasonal spike, that’s not an inconvenience. It’s an operational crisis.

This is the failure mode that gets reported upward. It shows up in dashboards, generates incident reviews, and resolves when the agent comes back online. It’s visible, it’s measurable, and it has a clear path to resolution.

Why the brand damage outlasts the outage

Thirty-four percent cite reputational damage and loss of customer trust, essentially tied with support overload. That near-tie is one of the most underreported findings in the survey, because these two failure modes don’t resolve the same way. The support queue clears. Brand damage doesn’t have a clear path back.

From the customer’s perspective, there’s no platform, no vendor, no infrastructure layer. There’s only the company’s brand. For 31% of organizations, the leading cause of a governance failure rollback is customer data exposure: personal information surfacing in an interaction where it shouldn’t have. That attribution is permanent in a way that a queue spike is not.

What makes this harder to address is that it often isn’t visible to the people who could act on it. Technical leaders report rollbacks at a higher rate than their business counterparts at the same organizations, 77% versus 69%. In retail, C-suite executives are 2.3 times more likely than their VPs and directors to say most AI communications pilots are succeeding. Same organization, very different accounts of the same events. That visibility gap is where the brand takes the hit.

The hidden engineering cost behind every AI launch

There’s a third cost that appears in neither the dashboard nor the customer complaint. Sinch data shows 84% of AI communications engineering teams spend at least half their time building guardrails and safety controls, instead of building the next customer experience. Thirty-five percent spend most of their time there.

And the direction of that burden surprises people. Production-stage engineering teams are spending more time on safety infrastructure than pre-production teams, not less. Each new agent, each new channel, each new compliance requirement adds another layer. The guardrail tax doesn’t amortize. It compounds.

“Every team needs to decide what controls belong at the platform layer and what their engineers should build on top, because the cost of building custom guardrails compounds over time, especially as the team moves through the product lifecycle,” says Anton Efimenko, SVP software engineering at Sinch. “Each new agent, each new channel, each new deployment adds to the pile. And eventually you lose that momentum when it comes to outperforming on the market.”

The real problem runs deeper than the AI itself

Across every statistical method applied to this dataset—correlations, regression models, cross-tabulations—one variable consistently outperforms all others as a predictor of AI deployment success: communications infrastructure satisfaction.

It’s not investment level, AI maturity, how long you’ve been in production, or how sophisticated your safety policies are.

The correlation between infrastructure satisfaction and AI deployment confidence is 0.52, the strongest relationship across 4,656 variable pairs analyzed in the study. How an organization feels about its communications infrastructure is a better predictor of AI success than anything else measured in the study.

Yet most organizations identify at least one significant shortcoming in their current provider. The most common gaps: insufficient reliability for AI at scale (42%), limited multi-channel capability (37%), and lack of AI platform integrations (32%).

And more than half of enterprises (55%) are custom-engineering the ability to preserve customer context when someone moves from one channel to another, from chat to voice, from WhatsApp to a phone call, because their platform doesn’t provide it natively. When a customer has to repeat themselves to an AI agent, they’re not experiencing a model failure. They’re experiencing the infrastructure gap directly. And it’s the company’s brand that pays the price.

The industry has voted with its budgets, trust, security, and compliance is the number one spending category globally, ahead of AI development itself. But most of that investment is going into application-layer guardrails built by engineering teams, treating symptoms while the infrastructure underneath stays the same. That’s why 74% are still rolling back agents. Companies can invest heavily in safety and still fail, because the failure modes originate one layer below.

Companies are already looking for alternatives

Enterprises haven’t fully articulated that diagnosis yet, but their behavior suggests they’ve felt it. Eighty-six percent have had active or exploratory conversations with alternative providers in the past 12 months, and only 4% have no plans to evaluate.

The strongest trigger for switching is experience. Ninety-one percent of enterprises that have had to roll back a live agent have evaluated or are actively evaluating a new communications provider. The most sophisticated buyers are the most active shoppers, not because they’re unhappy with a vendor, but because their AI ambitions have outgrown what the current infrastructure was built to handle.

When companies assess alternatives, reliability ranks first with 29% of respondents placing it at the top, ahead of compliance capability, ease of integration, and, notably, pricing. Pricing ranked eighth out of nine factors in the survey.

What this means for the next time you need help

Sixty-two percent of organizations have an AI customer communications agent live, and 88% will have one by the end of 2026.

Getting to production was hard, and most enterprises have made it. But the data is clear: Escaping pilot purgatory wasn’t the hardest part. Many organizations have deployed, they’re scaling, and what they’ve found on the other side is not what the market expected.

For the consumer on the other end of these interactions, the gap is immediate. When an AI agent fails mid-conversation, it often reverts to a human support team, one that was sized for a world where the AI was handling most of the volume. The wait gets longer, the frustration grows, and the brand takes a hit that doesn’t automatically resolve when the system comes back online.

The companies truly pulling ahead in this study aren’t just the fastest to deploy. They’re the ones whose AI stays live long enough to keep improving, backed by communications infrastructure that was actually built for the job.

This story was produced by Sinch and reviewed and distributed by Stacker.

The post Why 74% of AI customer service chatbots are pulled offline after launch appeared first on The Media Copilot.

For the first time in the internet’s history, bots account for more web traffic than humans.

Cloudflare CEO Matthew Prince announced the milestone this week, according to Tom’s Hardware, noting that automated traffic has now eclipsed human-generated requests online, months ahead of even his own projections.

“Welp, that happened faster than I predicted,” Prince wrote on X. “Thought it would be end of 2027, then early 2027, but agentic traffic growing so fast that bots have now passed human traffic online for the first time in the Internet’s history.”

According to Cloudflare’s Radar data, bots represented roughly 57% of all HTTP requests as of late April 2026, with humans accounting for the remaining 43%. Bot traffic has held between 53% and 60% in the weeks since. Prince said the actual crossover occurred in the last few months, though the data is messy enough that pinning down an exact date is difficult.

The shift underscores how quickly AI agents have transformed web traffic patterns. Before the generative AI era, bot traffic sat at around 20% of all web activity, with Google’s web crawler serving as the largest single source. Now, AI agents performing tasks on behalf of users are generating requests at a scale that dwarfs human browsing behavior.

Prince illustrated the contrast at SXSW earlier this year: “If a human were doing a task—let’s say you were shopping for a digital camera—you might go to five websites. Your agent or the bot that’s doing that will often go to 1,000 times the number of sites that an actual human would visit. So it might go to 5,000 sites. And that’s real traffic, and that’s real load, which everyone is having to deal with and take into account.”

The reaction to Prince’s announcement was swift. Tech billionaire Elon Musk replied with a single “Wow” to the post.

The full picture is more nuanced. While bots now dominate HTML request traffic—reading pages, scraping content, indexing sites—humans still account for roughly 65% of total web activity when the metric expands to include app usage, video streaming, maps, and social media scrolling. Bots have overtaken humans in the specific act of navigating and reading the web, but not in the broader measure of people actually using the internet.

Cloudflare, which handles approximately one-fifth of all global web traffic, has been tracking the trend closely. The company’s 2026 Threat Intelligence Report also found that bots now account for 94% of all login attempts across its network, meaning only 6% of login attempts come from actual humans.

The crossing point Prince initially forecast for 2027 arrived in 2026. What once required a two-year runway happened in a matter of months.

The post Cloudflare CEO: Bots have overtaken human traffic online appeared first on The Media Copilot.

Sitecore is acquiring generative engine optimization (GEO) startup Scrunch for around $225 million, according to a Bloomberg report, adding AI answer-engine visibility to an enterprise content platform that has been quietly building toward a machine-readable web strategy.

Neither Sitecore nor Scrunch have confirmed the price. The deal marks one of the larger investments in the emerging GEO market: the practice of optimizing brand content so it surfaces in AI-generated answers rather than traditional search results.

Scrunch’s platform shows brands real-time signals about how they appear across various AI platforms, along with competitive analysis and technical audits. Its Agent Experience Platform, or AXP, is designed to deliver content in formats AI agents can read and use without disrupting the human experience. Notable clients include Lenovo, Skims, Headspace, and Penn State University.

“We’re at a pivotal moment where companies must rethink traditional digital strategies and accept that the internet must be written for machines to understand if we want humans to experience it,” Eric Stine, Sitecore’s CEO, said in a statement.

Scrunch CEO and cofounder Chris Andrew echoed the same urgency in his own statement. “By joining forces, we’re helping companies meet buyers where they are, moving beyond traditional SEO to win inside AI-generated answers,” he said. “That’s where Scrunch’s AXP is a critical advantage, delivering content in a format AI agents can read and use, without disrupting the human experience, allowing brands to become the trusted sources that power those answers.”

The GEO space is becoming increasingly competitive as brands seek visibility in the AI experiences where consumers are spending more time. Scrunch previously raised $26 million, including a $15 million Series A last summer led by Decibel, with participation from Mayfield, Homebrew, and others.

The deal logic is in the numbers. Scrunch told ADWEEK last year that conversion rates in AI search are three to five times higher than in traditional online search, citing its own data. “A visitor coming from AI search is buying faster than a traditional organic visitor,” Andrew said at the time. Independent verification of those figures was not provided.

Third-party research offers some corroboration. In research conducted by Akamai, AXP-enabled webpages saw a 364% lift in brand presence in responses to non-branded AI prompts and a 218% spike in citations appearing in AI responses.

Stine said the combination would allow brands to “show up with greater clarity, authority, and relevance so they can build trust, increase share of voice, and influence decisions early in the buying journey when it matters most.”

The post Sitecore acquires GEO startup Scrunch for around $225 million appeared first on The Media Copilot.