Anthropic & Pentagon Resume Talks — Amodei Pushes for Mutual Agreement
Dario Amodei is back at the negotiating table with Emil Michael, the Under-Secretary of Defense for Research and Engineering, according to the Financial Times. In a CBS News interview, Amodei said Anthropic is trying to "deescalate" the standoff and reach "some agreement that works for us and works for them." The CEO framed the dispute not as a refusal to work with the military, but as a demand for basic assurances around autonomous weapons and mass surveillance.
In a memo to staff, Amodei revealed a striking detail: near the end of the original negotiations, the Pentagon offered to accept Anthropic's full terms if Anthropic deleted one specific phrase about "analysis of bulk acquired data" — the exact scenario Anthropic was most worried about. The company declined. Whether today's resumed talks produce a workable deal remains to be seen, but the tone has shifted from confrontation to negotiation.
Web Search, Web Fetch & Code Execution Now Generally Available
Anthropic has removed beta headers from three major API capabilities: web search, web fetch, and programmatic tool calling are all now generally available. Developers no longer need to include the beta flag to access these tools. The update also introduces dynamic filtering support via the new web_search_20260209 and web_fetch_20260209 tool versions — the model now writes and executes code to filter search results before they hit the context window, cutting down on token noise and improving response quality. Code execution is free when bundled with web search or web fetch.
Claude Code Gets /simplify, /batch, Memory Auto-Save & /copy Picker
A fresh Claude Code release ships several quality-of-life upgrades: two new slash commands (/simplify and /batch), memory auto-save so context persists across sessions without manual intervention, and a /copy picker that lets you grab specific code blocks from a response without selecting text manually. Smarter per-subcommand bash prefixes reduce false positives when Claude decides whether to run a shell command, and multi-agent memory handling has been reworked to prevent context bleed between parallel agents. OAuth token refresh, shell error recovery, and config integrity across concurrent instances all got bug fixes too.
Claude Sonnet 4.6 Now Generally Available in GitHub Copilot
Claude Sonnet 4.6 — launched on February 17 with a 1M-token context window, improved agentic search, and lower token consumption — is now fully rolled out in GitHub Copilot. Developers using Copilot Chat or inline completions can select Sonnet 4.6 as their model. The integration deepens the Anthropic-GitHub partnership and puts Claude's latest balanced model in front of millions of developers who may not have tried the standalone Claude products yet.
OpenAI Admits Pentagon Deal Was "Opportunistic and Sloppy" — Now Renegotiating
Sam Altman publicly acknowledged that OpenAI's Pentagon deal, announced the same week the White House blacklisted Anthropic, was rushed. He told reporters the company "shouldn't have rushed" it and called the original agreement "opportunistic and sloppy." OpenAI is now renegotiating the terms. The reversal tracks closely with Anthropic's position — and with the wave of user backlash that followed OpenAI's announcement, including the 295% spike in US ChatGPT uninstalls. Altman's walk-back is a meaningful validation of Anthropic's stance, even if neither company will frame it that way.
Defense Contractors Told to Drop Claude After Pentagon Blacklist
CNBC reports that defense tech companies are directing employees to stop using Claude immediately following the Pentagon's supply-chain risk designation. The move has real commercial teeth — Anthropic's products are being pulled from workflows at companies that can't risk their DoD contracts. The irony is sharp: Claude's usage in national security contexts was reportedly already extensive, including in Iran campaign targeting operations reported by the Washington Post. The Pentagon is now trying to expel the same tool it was actively using.
The Pentagon Standoff Is Becoming a Template for AI Safety in Contracts
What started as a contract dispute is turning into an industry-wide reckoning. OpenAI is renegotiating. The Pentagon is back at the table with Anthropic. Legal scholars say the supply-chain risk designation is on shaky ground. And Claude downloads are still running at record highs. The pattern emerging here is new: an AI company drew a public line around autonomous weapons and surveillance, paid a short-term commercial price, and then watched the rest of the industry move toward its position.
The interesting question now is whether resumed talks produce a deal with real teeth — one that actually prevents Claude from being used in fully autonomous targeting — or whether it's a face-saving reframe with the same substance. Anthropic's insistence on keeping the "bulk acquired data" language in any contract suggests they're not willing to paper over the core issue. The next 48 hours will tell a lot.