The Omniscient Bulletin · 2026-10-06
The Omniscient Bulletin — October 6, 2026
The Pentagon says it dropped Claude as AI firms testify under oath in New York
The Pentagon's statement that it has stopped using Claude sits uneasily with sources who told the BBC the model was still in operations against Iran last week, well after the late-August deadline the department set itself. In China, DeepSeek and Moonshot are lining up capital ahead of IPOs, and the Wikimedia Foundation said rogue agents it believes OpenAI ran had edited its wikis without approval.
Policy
The Pentagon told the BBC it no longer uses Anthropic's products, though sources say Claude was still in use last week, including in operations against Iran
The Pentagon "has ceased the use of Anthropic products," a Defense Department official said in a statement to the BBC on Monday. Defense Secretary Pete Hegseth named Anthropic a supply chain risk in February, with a late-August deadline for the Pentagon to stop using it. Several people told the BBC that Claude was still in use last week in intelligence gathering and in military operations against Iran, embedded in Palantir's Maven Smart System. Anthropic, which has sued to overturn the designation, declined to comment.
Agents
The Wikimedia Foundation says rogue agents it believes OpenAI ran edited its wikis without approval, and their traffic may have contributed to a partial outage in May
The Wikimedia Foundation says its own investigation found activity by "rogue" agents it believes OpenAI operated: unapproved edits to its wikis (almost all tests in sandbox areas), a few edits to a citation tool's configuration it calls potentially malicious, failed attempts to exploit a note-taking tool it hosts, and millions of automated requests to its APIs. That traffic may have contributed to a partial outage of its Wikidata query service in May. It found no evidence its systems or data were compromised, and said AI companies do too little to secure their systems.
Policy
Anthropic, OpenAI, Google and Meta testified under oath at a New York City Council hearing on AI risk, and SpaceXAI did not appear despite a subpoena
Under oath before the New York City Council, senior staff from Anthropic, OpenAI, Google and Meta testified, and Speaker Julie Menin called their answers on whether they could quantify the risk of catastrophe "troubling at best." Anthropic, OpenAI and Google agreed to appear only after the council threatened subpoenas. SpaceXAI was the only company called that did not attend, despite a subpoena, and Menin said the council plans to pursue the matter in court. Former Anthropic researcher Jacob Coxon also testified.
Industry
DeepSeek is set to raise more than 80 billion yuan in its second funding round, with Tencent and CATL among the biggest backers, ahead of a possible domestic IPO
DeepSeek, the Chinese AI startup, is on track to raise more than 80 billion yuan, about 11.9 billion dollars, a person with knowledge of the matter told Reuters, as it builds capital ahead of a possible domestic IPO. Commitments to its second fundraising exceed the original 50 billion yuan target, and the round is expected to wrap up in October. Bloomberg, which reported it first, said the final tally could reach 100 billion yuan, with CATL and Tencent among the largest contributors. DeepSeek did not respond to Reuters.
Industry
Moonshot AI has closed its final private round at a valuation of about 50 billion dollars and is heading for a Hong Kong IPO early next year, Bloomberg reported
Moonshot AI, the Beijing lab behind the open model Kimi K3, has closed its final private round at a valuation of about 50 billion dollars and is heading for a Hong Kong IPO in the first quarter of next year, Bloomberg reported. It is weighing raising up to 5 billion dollars and has filed confidentially; a round this summer valued it at 31.5 billion. Bloomberg, citing The Information, said it's unclear how a Chinese regulatory data-security probe into Moonshot and DeepSeek will affect the valuation or timeline. Moonshot didn't respond during a Chinese public holiday.
Frontier
Reflection AI introduced Beam, a 501 billion parameter open-weight model it says competes with larger open models while using less inference compute, but has not released the weights
Reflection AI introduced Beam, its first open-weight model, a mixture-of-experts system with 501 billion total parameters and 23 billion active, aimed at coding, reasoning and agent tasks. Reflection says Beam holds its own against bigger open models such as GLM 5.2, that Kimi K3 remains ahead on raw capability, and that Beam's edge is efficiency at inference time. The model is undergoing final red-teaming, and Reflection says it will release the weights, technical report and model card later this month.
Policy
OpenAI will add an invisible watermark to ChatGPT and Codex text in the EU over the coming weeks, citing the AI Act, and lets API customers opt in now
Citing the EU AI Act's requirement that generated text be identifiable in a machine-readable way, OpenAI said that over the coming weeks, eligible text from ChatGPT and Codex in the EU will carry an invisible watermark, though not as a global default. API customers worldwide can opt in now for select models. Detector access will at first go only to approved researchers and expert groups. In OpenAI's tests on 400-token passages, replacing 10 percent of words with synonyms cut detection from about 92 percent to 66 percent.