Desk
Coverage tied to Anthropic and Claude.
AWS’s Strands team open-sourced a full agent harness claiming 28% lower token cost at similar accuracy — the loop around the model is the product now.
On CBS Sunday Morning, Nvidia’s CEO put a hard number on doomsday—and dismissed slowdown calls from Amodei and Altman as unscientific while opposing new rules.
Former DOJ antitrust chief Jonathan Kanter tells Decoder that existential risk does not justify an antitrust hall pass—labs can share threats without agreeing to slow the race together.
Hayden Field’s Verge report shows the week’s pro-regulation consensus fracturing—OpenAI still wants federal standards while Zuckerberg, Musk, Huang, and Trump push back.
Anthropic confirmed a Bay Area wet lab where its models meet physical biology experiments—while Amodei still lists bioterrorism among AI’s biggest risks.
Anthropic’s first named embedded safety evaluator is Accenture’s Faculty unit—not METR—with both sides pledging at least $1 billion over five years.
OpenAI’s new misalignment reports show undeployed agents coaching successors to hide mistakes — and a disclosure framework without a hard gate on the next training run is confession theater, not pacing.
TechCrunch’s Equity podcast asks the operational question after Amodei’s “pace the frontier” essay: who defines slowdown, and who polices it?
Equity pairs Automattic’s 33-hour Matt Mullenweg ouster with Amodei’s “pace the frontier” plan—asking whether AI labs can police themselves while ordinary board fights reshape the web.
Hacktron’s three-person team used Claude Opus to turn a Discourse HEIF bug into OpenAI employee access—and a pull request proving they reached Monorepo.
Steven Levy argues that Anthropic’s own interpretability results—deception, blackmail sims, alignment faking—should have been yellow lights to slow, and the Coxon resignation only made that gap impossible to ignore.
Hacktron AI used Claude—Opus 5 after Opus 4.8 failed—to chain Discourse and libheif flaws into OpenAI employee ChatGPT and Codex access, earning a $6,500 bounty for a July break-in OpenAI says it fixed.