Anthropic vs U.S. government: procurement bans, supply chain risk, and lawsuits escalate with court rejections [climaxing]
Key Questions
What recent court decisions have impacted Anthropic's relationship with the U.S. government?
Federal appeals courts rejected Anthropic's bids to lift DOJ and Pentagon blacklists in April 2026, citing wartime needs. This escalates ongoing lawsuits over procurement bans and supply chain risk labels.
Why has the Pentagon blacklisted Anthropic?
The Pentagon labeled Anthropic a supply chain risk, leading to procurement bans. Appeals courts upheld this, denying Anthropic's requests for relief amid disputes with the Trump administration.
What does the Sabotage Risk Report say about Claude Opus 4.6?
The report on Claude Opus 4.6, used heavily for coding and agentic tasks at Anthropic, finds no dangerous goals. However, its BridgeBench hallucination accuracy dropped from 83% to 68%.
Why was the Mythos Preview withheld by Anthropic?
Anthropic withheld the Mythos Preview due to cyber risks, deeming the model too dangerous for public release. It raises national security concerns over high-tech risks like zero-days.
What are the highlights from the Gods/Monsters report?
The Gods/Monsters report highlights zero-days and Big Tech sharing related to AI risks. It ties into broader concerns about Anthropic's models and cybersecurity threats.
How did the UK AISI handle Anthropic's steering?
The UK AISI Model Transparency team replicated Anthropic's steering effectively. This demonstrates successful independent verification of Anthropic's AI safety techniques.
What upcoming events are related to Anthropic's disputes?
May hearings are pending amid protests and NIST involvement. These follow court rejections and ongoing escalations with the U.S. government.
What other recent Anthropic developments involve safety and partnerships?
Anthropic appointed Vas Narasimhan to its Board via the Long-Term Benefit Trust. Companies like Anthropic, Google, and Microsoft paid quiet AI bug bounties, including issues with GitHub agents.
Federal appeals reject Anthropic bids to lift DOJ/Pentagon blacklists (Apr 2026), citing wartime needs; Claude Opus 4.6 sabotage report shows no dangerous goals but BridgeBench hallucination drops; Mythos Preview withheld over cyber risks, Gods/Monsters highlights zero-days and Big Tech sharing; UK AISI steering replicates effectively; May hearings pending amid protests/NIST.