As AI companies continue scraping the web for training data, some publishers are experimenting with a new tactic that doesn’t only block unwanted bots, but tries to waste their time and money.

Known as LLM honeypotting, Digiday describes the approach as a form of deception to lure AI crawlers into consuming plausible-looking but ultimately worthless content, aimed at making large-scale data collecting so computationally expensive that it becomes less economically viable. 

Who’s doing this? A small number of publishers and e-commerce companies are looking for alternatives to traditional bot-blocking. And while the technique remains early and experimental, it’s one way for media companies to protect their content amid a fight to create standards for how AI companies track, value and compensate journalism. 

A cybersecurity tactic adapted for AI

Cyberhoneypotting is a long-standing cybersecurity strategy. The technique is to build a decoy target for attackers—or LLM bots, in this case—that can lure bad actors away and even gather intelligence on their capabilities and methods. In this case, the goal is to make scraping content without compensation more costly than it’s worth. 

Simon Wistow, co-founder of content delivery network provider Fastly, describes the philosophy as one of “Chang[ing] the economics of attacking.” If abusing a system becomes significantly more expensive than the value gained, the entire model will become unsustainable, he argues. 

Applied to AI crawlers, the strategy is equipped against all automated visitors, regardless of whether they are operated by major AI companies or smaller third-party scraping firms. 

Publishers can implement the tactic in several ways. They can introduce subtle delays, difficult (for computers) problems to solve before admittance, or endless mazes of contents and files filled with meaningless AI-generated text. Some honeypotting techniques go even further and try to inject bad data into the bots’ training datasets, “poisoning” the AI results.

The tactic’s adoption

Large e-commerce brands are already testing the technology successfully, said Wistow. News publishers are also showing increased interest, although he declined to identify specific customers. 

Even so, skepticism towards the strategy remains. 

Frederick Jahn, co-founder of AI company Centennal, argues sophisticated scrapers can often detect or avoid honeypots altogether. 

“I think it’s a good concept, but more on a marketing level, and like a gimmick,” Jahn said. He argues that publishers would be better served by creating real barriers to stealth crawlers, who are often not shown maze pages. 

Supporters of honeypotting maintain that, even if most scrapers adapt, increasing operational costs across thousands or millions of requests could make smaller scraping businesses financially unsustainable. 

“If they could burn through that 10 million funding in one crawl then suddenly those businesses aren’t viable and suddenly the whole market collapses, and that’s kind of what you’re going for,” said Wistow.

Costs and limitations for publishers

The strategy isn’t free. Generating and serving millions of fake pages is more expensive than simply blocking unwanted traffic. And larger publishers with more resources are better able to plan and implement the strategy.

Wistow said the approach is unlikely to become widespread, in part because of the consequences of filling the internet with yet more intentionally deceptive content. 

“Hallucinations happen even with good data, just because of the way LLMs work,” said Wistow. “This is about changing the economics for the people abusing your site, not running some giant disinformation campaign.”

The post Publishers Turn to AI ‘Honeypots’ to Fight Content Scraping appeared first on The Media Copilot.

OpenAI said that it plans to give a select group of government-approved partners early access to GPT-5.6 Sol, its most powerful AI model to date, before releasing the product more broadly. The limited rollout follows a request from the Trump administration, which asked the company to adopt a phased launch strategy for its next-generation AI system while security reviews are conducted. 

Last week’s request came from the White House’s Office of the National Cyber Director and Office of Science and Technology Policy, which have pushed AI developers to give federal agencies early access to frontier models so officials can evaluate their capabilities and potential security risks before wider deployment,

OpenAI CEO Sam Altman told employees in a memo that GPT 5.6-Sol will initially be available to 20 approved partners, including Amazon’s Bedrock platform. According to the memo, access is being granted on a “customer-by-customer” basis while the review process is underway. 

“We’ve made clear to the U.S. government that this is not our preferred long term model, and will work with them and others in industry to achieve a more sustainable approach for future releases,” Altman said in the memo. Altman said he hopes to release GPT-5.6 to the public a “couple of weeks later.”

The decision reflects growing concern over what the Trump administration says are legitimate national security implications of increasingly capable AI systems. 

OpenAI says GPT-5.6 Sol is its most advanced model to date, with improvements in reasoning, autonomous task execution, software engineering and cybersecurity-related capabilities. It released benchmarks that said its performance was broadly comparable to Anthropic’s Mythos 5, which was withdrawn on June 12 following a directive from the Commerce Department expressing concerns that its advanced capabilities could create new cybersecurity threats. 

At the time, Anthropic said the concerns were over “a small number of previously known, minor vulnerabilities” and that “other publicly-available models are able to discover them as well without requiring a bypass.” 

Politico reported that the initial vulnerability was brought directly to the White House by Amazon CEO Andy Jassy. (Amazon is an investor in Anthropic.)

“We have reviewed a report that we believe is the basis of the government’s directive and validated that the level of capability displayed there is widely available from other models (including OpenAI’s GPT-5.5), and is used every day by the defenders who keep systems safe,” Anthropic said in its statement.

Anthropic and the Trump administration have been sparring for months since a dispute over how Anthropic’s models would be used by the Pentagon. Reports that the company’s models were used in U.S. operations in Venezuela in January, in violation of its licensing terms, which prohibit using Claude models to commit violence, led to a near complete rupture. 

Adding further layers to the drama, just hours after OpenAI announced the limited rollout of GPT-5.6, Anthropic disclosed that the Trump administration had approved a limited release of Mythos 5, reversing the Commerce Department restriction. 

The restrictions on GPT-5.6 Sol and Mythos 5 follow a recent call from the Five Eyes alliance for closer coordination on advanced AI development and security. A White House official said the administration continues “to collaborate with frontier AI labs to develop shared approaches for addressing the challenges of scaling this technology.” 

The post Trump administration allows limited GPT-5.6 release appeared first on The Media Copilot.