Anthropic has given evaluators at the nonprofit METR employee-level access, and METR will publish its own assessments of the company's risks, incidents and practices. Anthropic has given up the right to block those reports.
Wired reports that attacks carried out with powerful AI models keep causing damage, yet victims are rarely pursuing the AI companies. Instead, they are being encouraged to adopt the same tools.
IBM argues that as AI agents take on more work, companies need an operating model linking agent actions to customer outcomes, infrastructure and financial accountability. Its approach folds automation, data and operational signals into one framework.
One of the photos President Trump posted on Truth Social to show White House renovations appears to have been altered with AI. The image of a "restored" column shows several inconsistencies and was flagged by Google's AI detector.
A WatchGuard Technologies survey of 684 employees at mid-sized companies in eight countries, including France, found that 64% use AI tools not approved by management. Fewer than 30% of the organisations keep an inventory of such software.
Decompiling games used to be a slow and very difficult task. A reverse-engineering community has closed because AI now does it with ease, 404 Media reports.