Akamai Lands an $11.6B Anthropic Deal, Stock Jumps 20%+
Akamai announced a seven-year, $11.6 billion agreement to supply Anthropic with CPU computing capacity across its distributed network of points of presence, with room to grow to roughly $9 billion more — a total possible value near $20 billion. Akamai shares jumped more than 20% in after-hours trading on the news.
The structure is notable: Anthropic receives warrants for nonvoting convertible preferred stock convertible into up to 5% of Akamai’s common shares at a $111.33 strike price, with about 2% vesting immediately and the rest vesting as Anthropic spends more over the term (roughly 1% per additional $3B committed). Akamai says it expects to spend about $5.5B in capex plus $1.7B to prepurchase supply-chain components like memory. CEO Tom Leighton called it a bid to be “the infrastructure provider for secure and responsible AI applications and workloads.”
White House Asks OpenAI and Anthropic to Slow-Walk UK Access
The White House has asked OpenAI and Anthropic to hold off giving UK AI Safety Institute (AISI) testers early access to new models until a US security review is complete. The stated reasoning: officials want to assess cybersecurity risk after incidents where AI agents gained unauthorized access to real systems, including an OpenAI agent that breached an Australian government health data portal in June.
No timeline for the review has been made public, and the White House, OpenAI and Anthropic all declined to comment when asked by Reuters. It’s a notable tightening for a UK institute that Anthropic and OpenAI have both previously given pre-release model access as a safety-testing courtesy.
Claude Opus 5.5 Is Now Live on AWS Bedrock
Claude Opus 5.5 — which shipped earlier this week at roughly 40% lower token prices than Opus 5 while matching Fable-class performance on most tasks — is now generally available on Amazon Bedrock. That puts it inside the same AWS IAM, VPC and enterprise-compliance tooling teams already use for other Claude models, no separate procurement needed.
For teams already running Claude Code against Bedrock (2.1.281 shipped Bedrock assume_role support and native guardrails earlier this week), Opus 5.5’s arrival closes the loop: cheaper frontier-class reasoning, same enterprise plumbing.
Vibe Check: Opus 5.5 Is Pulling Codex Converts Back
A round of hands-on developer reviews out this week found Opus 5.5 winning back people who’d switched to OpenAI’s Codex. Designer Tyler Nishida said his “jaw dropped at least five or six times this week” on visual-concept execution; Kieran Klaassen made it his daily driver, pegging it at roughly 90% of Fable’s coding capability with better collaborative creativity.
Not everyone’s fully back: Mike Taylor liked that Opus 5’s “annoying personality quirks” are gone but flagged weak time management on deadline-bound tasks, and Dan Shipper still leans 80/20 toward Codex, calling Opus 5.5 “a smaller Fable.” Testers also measured it as the most readable of the bunch by Flesch-Kincaid grade level, though it can bury the lede in longer responses.
Google, OpenAI and Anthropic Try to Hire Their Own Regulator
The three labs are reportedly courting Sriram Krishnan — the White House’s former senior AI policy adviser — to run a new, independent “Frontier AI Standards Agency” (also referred to as the Standards Authority for Frontier AI). The idea, which Google DeepMind’s Demis Hassabis has floated since around July, is a shared body that runs pre-release evaluations for dangerous capabilities using common terminology and testing methods across labs, with Hassabis previously suggesting a launch before the end of 2026.
No appointment is confirmed, and the group remains in working-group form. But three direct competitors jointly hiring a former administration AI czar to referee their own safety claims is a meaningful signal about where self-regulation is headed — and about who each lab wants credibility from.
Scientists Push Back: “CRISPR-Like” Isn’t CRISPR-Capable
Following Tuesday’s news that Claude’s biology lab flagged a new enzyme system (ART) with CRISPR-like repeat arrays, outside researchers are urging caution. Washington University microbiologist Kevin Blake: “There’s nothing to indicate this is a rival to CRISPR-the-technology, or could be developed into any kind of therapeutic or practical application.”
Computer scientist Dimitri Perrin drew the sharper distinction: “ART is CRISPR-like in its architecture, but there is no evidence that it is CRISPR-like in its function” — and noted the enzyme itself was already known; only the surrounding repeat-array pattern was new. One framing making the rounds: this could be apophenia, pattern-matching mistaken for discovery. Even Dario Amodei has said the “precise function, biotechnological utility (if any), or level of significance is not yet clear,” while Feng Zhang maintains it’s “genuinely intriguing” and worth investigating.
Anthropic's IPO Week Is a Study in Contrasts
Zoom out and today reads like a split screen. On one side: an $11.6B infrastructure deal and a serious bid to help write the industry’s own safety rulebook via the Frontier AI Standards Agency — both signals of a company acting like the confident, well-capitalized incumbent it’s about to prove it is on the public markets. On the other: a White House memo painting its CEO as an ideological threat to the “America First” agenda, and an administration quietly narrowing what its allies overseas get to see first.
None of these stories individually is decisive. Political memos circulate and get walked back; standards bodies get proposed and stall; compute deals get signed and take years to matter. But together they describe the actual terrain of an AI-lab IPO in 2026: it isn’t just a valuation exercise, it’s a simultaneous negotiation with capital markets, international regulators, domestic politics and the scientific community — all of whom are forming opinions on Anthropic in the same 24-hour news cycle, for reasons that have almost nothing to do with each other.