The same big tech companies that are stripping news publishers of site traffic are now dictating the terms of the emerging AI licensing market, and taking a significant cut in the process, according to a new report from the Open Markets Institute.

The report, “Same Gatekeepers, New Tollbooths: Mapping the AI Content Licensing Market,” argues that publishers are trapped in what the authors call a “double bind.” As Big Tech develops commercial AI products that siphon readers away from news sites, those same companies are positioning themselves as the gatekeepers of whatever alternative revenue streams emerge.

“Big Tech is occupying both sides of the value chain simultaneously,” write the authors, Courtney Radsch and Karina Montoya of the institute’s Center for Media & Digital Governance. They warn that “the deal structures, price precedents, intermediary take rates, and governance norms taking shape now will be difficult to revise once they are normalized.”

The report examines a growing ecosystem of AI content licensing marketplaces. Some are independent startups. Others are built by the very companies publishers are trying to negotiate with:

• ScalePost takes roughly 15% of revenue earned by rights holders.
• Cloudflare, which handles about 20% of global web traffic, takes an estimated 30% cut through its pay-per-crawl marketplace.
• ProRata.ai, which operates an answer engine built exclusively on licensed content, splits subscription and advertising revenue 50/50 with publishers. More than 500 publishers had signed up as of last summer.
• TollBit and Sphere.ai allow publishers to retain 100% of their revenue, instead charging AI companies a separate transaction fee.
• Microsoft’s Publisher Content Marketplace, announced in February, follows a pay-per-use model — but it’s not yet clear how much Microsoft will keep.

The report points to Spotify as a benchmark: despite a 30% take rate, the streaming model allowed music rights holders to earn significant revenue and stabilize the industry during a turbulent transition. The authors argue similar scrutiny is needed for AI licensing marketplaces, particularly when Big Tech is building the scaffolding.

“Regulatory attention is warranted on these platform operators in order to mitigate their data access advantages and ability to set de facto and potentially coercive standards for an industry in which no independent standards yet exist,” the authors write.

The post New report finds wide disparity in AI tollbooths for publishers appeared first on The Media Copilot.

Anthropic accidentally published the source code for Claude Code this morning — 512,000 lines of TypeScript, exposed for hours before anyone pulled it.

Source maps can reconstruct the full, readable source from compiled code. By 4:23 am ET, a developer at Solayer Labs had spotted it, posted a direct download link on X, and the codebase was being mirrored to GitHub and analyzed by thousands of developers across the industry.

Anthropic confirmed the incident to VentureBeat: “Earlier today, a Claude Code release included some internal source code. No sensitive customer data or credentials were involved or exposed. This was a release packaging issue caused by human error, not a security breach. We’re rolling out measures to prevent this from happening again.”

That framing — human error, not a breach — is accurate but undersells the competitive exposure. Claude Code is reportedly generating $2.5 billion in annualized recurring revenue, has more than doubled since the start of the year, and competes directly with Cursor, GitHub Copilot, and a fast-moving field of agentic coding tools. The leaked code gave competitors a detailed look at exactly how Anthropic solved some of the hardest engineering problems in that space.

Among the details developers surfaced: a three-layer memory architecture that keeps AI agents coherent across long coding sessions by maintaining a lightweight index of pointers rather than storing everything; an “autoDream” background process that consolidates the agent’s memory while the user is idle; 44 hidden feature flags; and internal model codenames including Capybara (a Claude 4.6 variant), Fennec (Opus 4.6), and an unreleased model called Numbat. The code also showed internal performance metrics: version 8 of Capybara had a 29-30% false claims rate, a regression from version 4’s 16.7%.

The most discussed detail was what developers called “Undercover Mode” — a system that lets Claude Code contribute to public open-source repositories without disclosing that the commits came from an AI. The leak is the second high-profile setback for Anthropic in recent weeks, following its lawsuit against the Pentagon over the company’s AI safety limits.

The underlying story here isn’t the leak itself — packaging errors happen. It’s that one slip exposed enough proprietary engineering detail that competitors now have a working map of how one of the most commercially successful AI coding tools actually functions. For a company that’s staked its position on responsible AI development and technical differentiation, that’s the real cost.

The post Anthropic accidentally published Claude Code’s source code this morning appeared first on The Media Copilot.

Microsoft wants to stop being your AI assistant and start being your AI workforce.

That’s the pitch behind Wave 3 of Microsoft 365 Copilot, announced Monday. The centerpiece is Copilot Cowork — a new capability, built with Anthropic’s technology, that lets Copilot take on long-running, multi-step tasks rather than just answer a single prompt. Think: not “draft this email” but “research our top five competitors, build a summary deck, and schedule a review meeting.”

Cowork is still a research preview, but it points at something real. Most enterprise AI disappointments so far have come from the gap between generating something useful and actually finishing something. Cowork is Microsoft’s attempt to bridge that.

Wave 3 also brings Copilot deeper into the apps people actually live in. Word, Excel, PowerPoint, Outlook and Copilot Chat are all getting what Microsoft is calling “next-generation agentic experiences” — the ability to create and revise documents, spreadsheets, and decks from inside the app, with everything saved into governed Microsoft 365 environments. That last part matters more than it sounds. Keeping AI-generated content inside existing permission and compliance systems is the real enterprise sell.

On the model side: Claude is now available directly in Copilot Chat for customers in the Frontier program, alongside OpenAI’s latest. Microsoft is positioning itself as model-agnostic — your AI interface, whoever’s model wins.

The governance piece is Agent 365, a $15-per-user control plane for IT and security teams to observe, manage and secure agents across the organization. Generally available May 1.

And then there’s the bundle. Microsoft 365 E7 — the new Frontier Suite — lands May 1 at $99 per user. It wraps Microsoft 365 E5, Copilot, and Agent 365 into one package, with Microsoft Entra, Defender, Intune and Purview security baked in. It’s clearly aimed at CIOs who want to buy the whole stack from one vendor and not think about it again.

Whether enterprises will pay for all of it is another question. But Microsoft is betting that once AI moves from chat to actual task completion, the governance and security layer becomes non-negotiable — and they want to own that layer.

The post Microsoft turns Microsoft 365 Copilot into a broader agentic work platform appeared first on The Media Copilot.

Adobe is widening its AI push on two fronts: new Photoshop and Firefly editing features under embargo, plus Adobe Express and Acrobat integrations inside Microsoft 365 Copilot.

Adobe is making a broader bet that creative work will increasingly start inside conversational AI and stay inside the apps people already use. That strategy showed up Monday in two linked announcements: new AI-powered editing capabilities in Photoshop and Firefly, plus a new Adobe foothold inside Microsoft 365 Copilot.

The company is rolling out a public beta of AI Assistant in Photoshop on web and mobile, alongside new Firefly image editing capabilities designed to let users make more complex edits through prompts, voice input, and step-by-step guided changes.

The pitch is straightforward: instead of forcing users to understand Photoshop’s full toolset before they can make useful changes, Adobe wants people to describe what they want done — remove distractions, adjust lighting, swap backgrounds, refine color — and let the software either execute or walk them through it. That is a meaningful shift for journalism teams, social editors, and audience staff who need fast visual iteration but do not always have a designer on hand.

Adobe is pairing that with a bigger distribution play. As part of Microsoft’s Wave 3 Microsoft 365 Copilot rollout, Adobe Express and Acrobat are being brought into Copilot chat for enterprise customers. Microsoft’s broader announcement frames the update as part of a push toward more agentic work inside Word, Excel, PowerPoint, Outlook, and chat-based workflows. Adobe’s add-ons extend that logic into content creation and document work.

In practical terms, that means a user inside Copilot could move from drafting an idea to generating or revising branded creative in Express, or from document chatter to creating and organizing PDFs in Acrobat, without bouncing across separate tools and browser tabs. For newsrooms and media companies, that matters less as a flashy product demo than as a workflow signal: the winners in AI productivity may be the tools that insert themselves into existing enterprise habits, not the ones that demand a separate destination.

That also gives Adobe a stronger answer to the platform risk that comes with the rise of ChatGPT, Copilot, and other assistant interfaces. The company has already pushed its apps into ChatGPT. Now it is doing the same inside Microsoft’s workplace stack, which is where many media companies already live.

There is still reason for skepticism. Prompt-based image editing is getting crowded fast, and “simpler creative workflows” is now a promise from nearly every major platform. What Adobe has going for it is incumbency: Photoshop remains the professional default, Acrobat still owns a huge share of document workflows, and Express gives it a lighter-weight product for non-designers. If the company can make those tools feel native inside Copilot rather than bolted on, that is more defensible than yet another standalone AI feature drop.

For publishers, the bigger takeaway is that generative AI is no longer arriving only through newsroom-specific tools. It is getting woven into the software stack employees already use for slides, spreadsheets, email, design, and document management. That raises the bar for governance and training, but it also lowers the friction for adoption.

Along with the Microsoft 365 Copilot integrations, they show where the company thinks the next creative battlefield is: not just inside Adobe apps, but inside the AI assistants that increasingly mediate everyday work.

The post Adobe expands Firefly and Photoshop AI while pushing into Microsoft 365 Copilot appeared first on The Media Copilot.

Content authentication standards exist. They work reasonably well in controlled conditions. The problem, according to a new Microsoft research report, is that almost nobody has deployed them at scale — and the window to fix that is closing fast.

The report, Media Integrity and Authentication: Status, Directions, and Futures, evaluates three authentication approaches — cryptographically signed provenance metadata (C2PA), imperceptible watermarking, and soft-hash fingerprinting — and finds that while the technology is mature enough, adoption remains fragmented across devices, editing tools and distribution platforms. Without broad implementation, the report warns, the gap between what AI can generate and what audiences can verify will keep widening.

The strongest finding: layering C2PA signing with imperceptible watermarking delivers “high-confidence provenance authentication” — a verifiable chain of custody from creation to publication. Fingerprinting is better suited for forensic work than real-time verification at scale.

The report introduces a concept that should concern newsrooms directly: “sociotechnical provenance attacks.” These aren’t just technical file manipulations. They exploit user perception — making real content look fake, or synthetic content look legitimate. Visible watermarks without cryptographic backing, the researchers warn, can actually make these attacks easier by training audiences to trust signals that can be forged.

There’s also a hardware problem. High-confidence authentication requires secure enclaves built into cameras and capture devices at the hardware level. Most devices don’t have this yet. Until they do, provenance claims on content captured with conventional equipment remain easier to dispute.

For media organizations, the takeaway is sobering: the C2PA ecosystem Microsoft helped co-found in 2021 now has thousands of members, but deployment in actual newsroom workflows and consumer platforms remains thin. The report frames 2026 as a critical inflection point, with regulations approaching and generative AI accelerating. Fragmented adoption now means fragmented trust later — exactly the environment where AI-driven misinformation is most effective.

The post Microsoft report: current media authentication tools aren’t ready for the AI content flood appeared first on The Media Copilot.

Microsoft announced its Publisher Content Marketplace on Feb. 4, a platform designed to broker licensing deals between AI companies and publishers. The marketplace lets publishers control how their content is licensed for AI training and receive payment based on actual usage.

The platform, called PCM, functions as a central hub where publishers license text, images and other media to AI developers under terms they set. Microsoft positions it as infrastructure for what it calls “the agentic web,” where AI agents will increasingly mediate information access.

The marketplace addresses a friction point in AI development: companies need training data, publishers want compensation, but negotiating individual deals is slow and opaque. PCM standardizes the process with usage tracking and per-use payment models.

Major publishers have already signed licensing deals outside this marketplace. News Corp struck agreements with both Google and OpenAI. The Associated Press, The Atlantic, Vox Media, Axel Springer, The Washington Post and TIME have all licensed content to AI companies in individual negotiations.

Microsoft’s marketplace changes the dynamic from bilateral negotiations to a platform model. Publishers post their content and terms. AI developers browse and license what they need. Microsoft handles the technical infrastructure and presumably takes a percentage, though the company has not disclosed marketplace fees.

The timing matters. Meta signed multiyear licensing deals with CNN, Fox News, USA Today, Le Monde Group and others in December 2025 to bring real-time news into its Meta AI assistant. These deals happened before Microsoft’s marketplace launched, suggesting appetite for systematic content licensing continues to grow.

For newsrooms, the marketplace represents another revenue option in a landscape where direct traffic from AI-powered search threatens existing business models. Digiday reported in December that publishers give Big Tech’s AI licensing deals mixed grades, with concerns about appearing in AI search products that cannibalize their own traffic channels.

The marketplace model could make licensing more accessible to smaller publishers who lack resources for complex contract negotiations. But questions remain about pricing power, usage verification and whether per-use payments will generate meaningful revenue compared to lump-sum deals some publishers have negotiated directly.

OpenAI reportedly plans to retire several models including GPT-4.1 in February 2026, according to Future Tools. That kind of model churn could complicate licensing agreements tied to specific AI systems rather than platform-level deals.

Microsoft’s marketplace is live now, starting with Copilot as the first AI builder using licensed content.

The debate over AI licensing comes as newsrooms grapple with whether to pursue litigation or negotiation with AI companies. Some publishers view licensing as a pragmatic revenue stream, while others worry about AI scrapers bypassing their protections entirely.

The post Microsoft launches marketplace to broker AI licensing deals between publishers and developers appeared first on The Media Copilot.