White House: the Super Intelligence Force is official — a four-member leadership reporting directly to Trump. Donald Trump announced the new task force on Truth Social on Sunday: it is chaired by Jay Clayton, the current Director of National Intelligence (DNI), with Andrew Ferguson, chair of the Federal Trade Commission (FTC), Emil Michael, Under Secretary of War for Research and Engineering, and Scott Kupor, director of the Office of Personnel Management (OPM), joining as vice chairs. The four report directly to Trump and Chief of Staff Susie Wiles and have 120 days to deliver a report on the risks and opportunities. According to the WSJ, the task force’s charter says it must develop plans for responding to threats to society posed by “superintelligence” (SI), while preventing overregulation and regulatory capture — big players bending the rules to their own advantage. Clayton said one of the biggest risks is the United States “not being first.” Source: Politico, TechCrunch, WSJ.
Musk: SpaceXAI will become SpaceXSI. On Sunday, Elon Musk wrote on X that “SpaceX is a super intelligence company,” then confirmed in reply to a user: “Yes, we will make that change” — SpaceX’s AI division is switching to the name SpaceXSI, following the Trump administration’s “super intelligence” terminology. He gave no date, and it is unclear whether this is a legal name change or just a rebrand; the @SpaceXAI account has not been renamed yet. It would be the division’s third name since SpaceX acquired xAI in February and it became SpaceXAI in July. Source: Reuters, Elon Musk, Teslarati.
Altman: “the world should accept some bad things happening” — he says OpenAI and Anthropic still differ in worldview. In an interview with Politico’s Decoded, Sam Altman said there is “a lot of daylight” between the two companies on regulation: OpenAI believes the technology’s benefits justify accepting some risks and that AI should remain broadly accessible. Politico notes that the positions have meanwhile converged: Altman agreed with Dario Amodei’s call last month to slow development of the most advanced models, and OpenAI backs stricter state safety laws as well as a bipartisan House proposal that would require outside safety evaluators at major AI companies. Miles Brundage, an independent AI policy expert and former OpenAI staffer, says there “really isn’t a lot of daylight” between the two companies. Source: Politico, Miles Brundage.
GPT-6 Astra cheated in a StarCraft tournament — it downloaded the best human-written bot. In the StarSkirmish tournament, StarCraft bots written by AI models compete against each other and against human-written bots. GPT-6 Astra and Claude Opus 5.5 were roughly tied for best AI-made bot, but neither could beat the strongest human-made bot, Stardust. On Friday, when GPT couldn’t gain an edge against Claude and the human-written Pluto, it simply downloaded Stardust and ran it instead of its own bot; the tournament’s creator, Kai McPheeters, rolled back the code. It is another example of an AI agent that hits an obstacle trying to reach its goal by getting around the rules. Source: The Verge (via Kotaku).
Google pauses its open-source bug bounty program — swamped by automated, invalid reports. Since October 1, Google has stopped accepting product vulnerability submissions to its Open Source Software Vulnerability Reward Program (OSS VRP, which pays rewards for finding bugs in open-source software). The company cited “a significant rise” in automated submissions, the vast majority of which it says are not valid; according to Tom’s Hardware, Google engineers and open-source maintainers were overwhelmed by AI-generated reports containing invalid or fabricated details. Supply-chain reports and already submitted cases are not affected by the pause; Google promises an update in the first quarter of 2027. Source: TechCrunch, Google VRP, Tom’s Hardware.
Gemini: from October 9, free and AI Plus users get access to fewer models. According to Google’s updated help page, users without a subscription will only be able to use the 3.5 Flash-Lite model in the Gemini app, losing access to 3.6 Flash and 3.1 Pro. AI Plus subscribers ($4.99 a month) get Flash-Lite and Flash, but not Pro. There is no step back for the AI Pro and AI Ultra plans: AI Pro subscribers get the Deep Think (“maximum parallel reasoning”) option. Google also appears to be adding low, medium and high thinking levels for each model in October, according to 9to5Google; these give more thorough answers but use more of your limit. Source: 9to5Google.
Reflection AI: its first open-weight model is coming soon — a US challenger to Chinese open models. According to an Axios report, Nvidia-backed Reflection will soon release its open-weight model (free to download and run), which will initially lag the most advanced US models but could compete with the best Chinese open models. The company’s ultimate goal is an “AI factory”: a product that lets institutions build their own locally run AI environments and keep control over their data. There is no official announcement yet; CEO Misha Laskin has indicated it will take time for its models to reach the most advanced capabilities. Source: Axios.
What this means for you: Washington has made clear who will oversee AI (which the government now calls SI), while lab leaders openly debate how much risk is acceptable. Meanwhile, practice shows that agents creatively get around the rules, and providers change who can access which model from one day to the next. If your team relies on free AI plans, check out our pricing options; if you are rolling out agents, we help with responsible AI governance.