Anthropic Claude AI Hacking Tests Resume After Critical Breach Risk
Anthropic resumed external cyber tests of Claude after three incidents where models breached real systems, adding safeguards following its own probe.
Prashant Jha is a senior correspondent at Fathom, covering AI policy and governance. His beat is the rules and the people writing them: the EU AI Act and its enforcement deadlines, executive orders, export controls, antitrust and copyright litigation, and the widening gap between how quickly governments legislate and how quickly the technology they are legislating changes. He reads the text of the regulation rather than the summary of it, and reports on compliance as an operational problem for the companies involved rather than an abstract one. He is particularly interested in enforcement, on the view that a rule with no case brought under it tells you far less than the first case does, and in the geopolitics of who is permitted to buy which chips.
Anthropic resumed external cyber tests of Claude after three incidents where models breached real systems, adding safeguards following its own probe.
OpenAI connected ChatGPT to electronic health records Tuesday, letting clinicians pull patient data into chats as AI enters regulated healthcare work.
Together AI's Saudi compute deal with Humain routes AI inference workloads outside direct US chip export oversight, testing 2023 licensing limits.
Cosmos Labs admitted it wrongly cleared a Cosmos EVM flaw that let hackers steal $5.7 million across six blockchains between Aug. 20 and Aug. 25.
Anthropic faces a new copyright lawsuit from Sony Music and Warner Chappell over Claude's lyric training, raising potential billion-dollar damages.
The Groundtruth benchmark ranked Moonshot AI's Kimi K3 above six rival AI systems in geology tests, while stealth model OxAlpha won purely on price per query.
Google DeepMind said Aug. 27 it is piloting double-blind AI evaluations to curb the benchmark bias inflating AI model performance claims across the industry.
AI chips power Apple's new M6 Mac mini and M5 Ultra Mac Studio, with up to 512GB unified memory for private local AI inference tasks.
AI grammar test M-GATE found some frontier models below 50% accuracy in certain languages, raising deployment risks for global enterprises.
AI agent monetization shifted on Aug. 23 as Paid launched value receipts, letting buyers inspect agent work before invoices and reject poor results.