Hawley and Murphy: AI companies liable when agents hack
Today's AI news splits between concrete and rhetorical: a $15 billion data center plan that actually has land and power attached, and a liability bill that has sponsors but no text yet.
Hawley and Murphy plan a bipartisan bill making AI companies liable for agent-driven hacking. The AI Agent Accountability Act would hold AI developers both civilly and criminally liable when a recklessly designed model is used to break into systems — the first bipartisan Senate move to put hard liability on the labs rather than regulate their behavior. Senator Josh Hawley (R-Mo.) sketched the plan at a Homeland Security subcommittee hearing he chairs on Tuesday: companies liable for reckless design, deployers liable for reckless deployment, and clarity that criminal hacking penalties reach AI companies and their agents; Senator Chris Murphy (D-Conn.) is the co-sponsor, per an Axios report. The design is deliberately narrower than the duty-of-care drafts that have circulated for months — instead of giving regulators a veto over a release, it borrows the burden from product liability and lets courts decide what reckless means (we weighed that broader approach last month — Senate weighs a duty-of-care law that could block unsafe AI models). Two details set the stakes: OpenAI's Sam Altman was invited to the hearing and declined, and the White House's line, delivered by DNI Jay Clayton, is that existing consumer and product law already covers all of this. Congress may be packing up for the election season, but a named R-plus-D pair arguing in public that the current law is not enough is the first real counterweight to the self-regulation pledge six CEOs signed on Tuesday.
Dell, Jera, and London's RHAELM plan a $15 billion, 400-megawatt AI data center next to a working gas station in Chiba. The Financial Times reports the project as one of the largest AI infrastructure builds in Asia outside China; Jera's own announcement fills in the mechanics — the site sits behind the meter at its Chiba thermal power station outside Tokyo, drawing up to 400 megawatts directly from the utility's existing generation, with total capital deployment across phases expected to exceed $15 billion and operations targeted to start around 2028. Roles are split cleanly: Jera supplies land and round-the-clock power, Dell brings standardized rack-scale compute as its AI Factory, RHAELM develops and operates, and Apollo has committed to serve as strategic investment and financing partner. Read it as a template more than a project. Grid interconnection is the bottleneck everywhere, and behind-the-meter power sidesteps the queue entirely — the AI buildout is increasingly sized by who controls generation, not who controls fiber. We looked at the demand side of this argument in AI data centers will out-burn Germany and Japan on gas by 2035.
Barclays is scaling Claude across the bank, with a target of half its developers on Claude Code by the end of this year. Anthropic and Barclays announced the expansion together: the bank expects Claude Code adoption to reach 50% of its developer population by the end of 2026 and a majority of software engineers in 2027, on top of deployments already running — a knowledge assistant used by more than 16,000 colleagues that has handled over one million searches, and a Global Markets email-routing platform processing roughly 120,000 messages a day. What to watch: the adoption targets are the real signal here, not the partnership language — enterprise AI deals are usually announced with numbers nobody has to hit, and these are dated commitments from one of the largest banks in the world.
Is civil and criminal liability for reckless model design the right line — or does it criminalize engineering judgment? Tell us in the comments.
Sources: Axios via Yahoo News · Roll Call · Financial Times · JERA announcement · Anthropic announcement · Techmeme
