Eigen RadarAI
Archive
Analysis

A closed briefing, an industry draft and a court order

A White House cyber framework briefed to major labs without published criteria, an Nvidia-led industry draft on incident sharing, and an appeals court order on Perplexity's shopping agent show AI rules forming outside open public rulebooks.

Artificial Intelligence··Morning
On a sunlit walnut bench, a frosted cylindrical housing, an open brass collar crowded with adjustment screws, and a raised spring latch converge on one dark ceramic object.

Labs were briefed on a secret cyber framework

WIRED reports that a White House official confirmed the administration has finalised its plan for the cybersecurity risks of capable AI models. Staffers from OpenAI, Anthropic, Google, Meta and Nvidia were briefed at the White House on Tuesday; testing criteria and covered models were not disclosed. Developers may submit new models to the federal government voluntarily up to 30 days before public release. The White House will vet cyber capabilities against a classified benchmarking system and share models with federal agencies and trusted corporate partners. WIRED, citing Axios, says open models are reportedly excluded. The framework stems from an executive order President Donald Trump signed earlier this year, which says the arrangement should not be treated as a mandatory licensing regime. A second White House official, speaking anonymously, said the framework is intentionally narrow and covers only the cybersecurity capabilities of the most advanced models, naming Anthropic's Fable and OpenAI's ChatGPT 5.6. Brad Carson of Americans for Responsible Innovation said the rules should be public; ControlAI's Conor Leahy argued regulation should remain non-voluntary. The White House did not respond to requests for comment. The report also recalls June temporary export controls on Anthropic's most advanced models and OpenAI's delay of the GPT-5.6 rollout at the White House's request.[1]

An industry alliance opens incident sharing to comment

Separately, the Shared AI Findings Exchange working group inside the Nvidia-led Open Secure AI Alliance has asked for public input while drafting guidelines on how AI cybersecurity incidents should be shared. TechCrunch reports that the initial proposals cover confidential reporting procedures, alerting affected parties and blame-free analysis. The alliance brings together more than 120 member companies under Linux Foundation management. Named members include Adobe, Amazon, BlackRock, Cisco, Hugging Face, Intel, Microsoft, Okta, Red Hat and Visa; Anthropic, Google and OpenAI are not members. Contributions listed include Nvidia's open models and Garak scanner, Okta's agent identity technology, Red Hat's agent governance tools, and Amazon's Strands Agents tool with the Cedar authorisation language. The proposals were presented at the Black Hat conference in Las Vegas. TechCrunch writes that the group formed after pressure over potential US restrictions on Chinese open-weight models, and relays the alliance's argument that openness may be an important path to AI safety.[2]

A shopping-agent injunction and three rule venues

On the same reporting day, the Ninth Circuit Court of Appeals reversed the preliminary injunction that barred the shopping tool in Perplexity's Comet browser from reaching Amazon's store. Engadget reports that the court concluded Amazon had not established a Computer Fraud and Abuse Act violation. The court found that Comet needs a user's direction to reach Amazon, so access is attributed to users rather than to Perplexity. Under the statute Amazon must prove intentional unauthorised access, information taken and damages averaging at least 5,000 dollars a year. The ruling states that an injunction against conduct that likely does not violate the CFAA would not serve the public interest. Amazon said it respectfully disagrees with the preliminary-injunction decision, remains confident in its case and is evaluating next steps. It may seek a rehearing or petition the Supreme Court; the underlying lawsuit continues in federal court in San Francisco. Together with the White House briefing and the industry comment process, the three developments place material rules for capable models and shopping agents in a voluntary federal cyber framework with unpublished testing criteria, an industry alliance that holds membership and draft procedures, and a court order under cyber-access law.[3], [1], [2]

References

  1. News sourceWIREDAI labs were briefed on the cyber framework; the criteria were not published↩1↩2
  2. News sourceTechCrunchThe Nvidia-led alliance opens its incident-sharing rules to comment↩1↩2
  3. News sourceEngadgetAn appeals court lifts the injunction on Perplexity's shopping agent↩
AI safety and regulationChip wars and the supply chainThe frontier model race

More analysis from this desk

View all Brief analyses from the Artificial Intelligence desk →