The newly minted Super Intelligence Task Force is telling AI companies to report and immediately remediate any instances of rogue behavior involving their products, a rare effort by the Trump administration to exert more power over the firms developing the advanced technology.
An Anthropic incident in which a testing model submitted visa applications, according to a State Department spokesman, spurred the task force’s first “action,” which it’s referring to as a “notification and remediation process.”
Frontier AI labs largely do this sort of disclosure and incident reporting already, and Trump’s task force has not laid out exactly what will count as an “incident” that must be reported going forward.
Still, the administration’s decision to make an example of Anthropic is likely to trigger more disclosures than before.
“Our message to all SI companies is clear: delayed notification, inadequate corrective action, and a failure to take responsibility will not be tolerated,” task force leaders Jay Clayton, Andrew Ferguson, Scott Kupor, and Emil Michael said in a statement.
The effectiveness and scope of the new requirements will depend in part on how the task force defines an “incident,” how quickly it must be disclosed, who it must be reported to, and the consequences for not doing so. Officials have yet to lay out these details, and companies are seeking more clarity.
Anthropic contacted the task force to disclose the latest incident and has a history of reporting security episodes publicly and tweaking models to prevent more harm.
— Ashley Gold