AI News from THE DECODER
Latest coverage from THE DECODER, summarized and scored for signal.
- Exit polls show AI ranked last among voter priorities in the 2026 U.S. midterms — NBC News and SSRS exit polls from the 2026 U.S. midterms put AI and data centers last among five issues, with 56 percent calling it decisive or important, showing public attention to AI has not yet become a top voting concern.
- Schools that turn off Google's AI will lose search too starting in 2027 — From early 2027, Google will switch off Gemini and Search by default for student accounts, but schools that want Search back can no longer get a version without AI, putting pressure on districts that already pulled back on AI.
- OpenAI says GPT-6 Astra designed a hardware device from scratch in under a week — OpenAI says GPT-6 Astra designed a Codex voice device end to end, from enclosure and PCB to firmware, in under a week, while an unnamed internal model scores 0.875 on BenchCAD 2.0 versus Astra's 0.594.
- Rapid LLM development is outrunning the research meant to explain it, study finds — Stanford researchers retested 50 widely cited LLM findings on 20 current models and found 44% have expired and 34% weakened, with an average lifespan of 15.7 months for dated claims.
- AI agents take over at SETI's Allen Telescope Array, scanning for extraterrestrial intelligence — The SETI Institute and Breakthrough Listen now use AI agents on Nvidia Holoscan to steer the Allen Telescope Array and analyze radio signals live, cutting weeks of work to hours.
- OpenAI hires ex-Army Secretary, gaining bipartisan reach just as midterms loom — OpenAI hired former Army Secretary Dan Driscoll as Senior Counselor for National Security Modernization, deepening its push for US defense contracts and a voice in military AI rules ahead of the midterms.
- Anthropic's Message Threads cut time to first token by 35 percent for long conversations — Anthropic released Message Threads in public beta for the Messages API, so agents send only what changed since the last turn, cutting time to first token by about 35 percent on 500K to 1M token requests.
- Nvidia pushes cloud customers to cut back on China business — Nvidia has told some cloud customers, including neoclouds it backs, to take on less business from Chinese companies, closing a gap where U.S. export rules cover chip shipments but not remote cloud access.
- Huawei tells customers to find their own AI chip memory, effectively outsourcing procurement to the black market — Huawei is telling some customers to source HBM for its AI chips themselves, pushing buyers toward smuggled Samsung and SK Hynix memory as U.S. export rules and weak domestic supply choke its Ascend roadmap.
- Bug hunters can now earn up to $172,000 from OpenAI as company ramps up security — OpenAI quadrupled its bug bounty payouts, with critical ChatGPT and Codex flaws now worth up to $30,000 and bonus-stacked totals reaching about $172,000, as it launches a new security sprint with the firm Cathedral.
- Anthropic cracks down on game cloning in Claude Code after wave of AI-built knockoffs — Anthropic has added rules to Claude Code's Opus 5.5 system prompt that block it from recreating existing video games, after users flooded the web with AI-built browser ports of titles like Halo and GTA Vice City.
- Goldman Sachs pulls AI coding tools from traders and bankers — Goldman Sachs has restricted Claude Code and GitHub Copilot to engineers only, reversing months of use by traders and bankers and signalling a pullback from non-technical vibe coding at big banks.
- AI elites might hoard intelligence like mercantilists hoarded gold — Engineer Venkatesh Rao argues that elites treat intelligence like mercantilists treated gold, as a scarce asset to control, and that cheap AI could instead raise demand for human thinking, so policy should focus on fair access rather than protecting legacy knowledge jobs.
- SpaceXSI apparently cuts Grok Bot quotas weeks after Musk called it his fastest-growing product — SpaceXSI appears to have halved Grok Bot usage quotas for some users, with Musk saying running costs were unsustainably high relative to revenue, less than two weeks after he called it his fastest-growing product.
- Anthropic grows revenue twelvefold but posts $23 billion loss — Anthropic posts ~$17B quarterly revenue (12× YoY) but a $23.3B net loss as compute lock-ins and donation matching erase the prior quarter’s profit ahead of a large IPO.
- One brutal Minecraft night scared 50 AI agents into inventing chicken handoff protocols, a Fertilizer Officer, and fishing shifts — YC Labs and Olam Labs put 50 GPT-6.1 Sol agents in Minecraft: after a deadly first night they built extreme bureaucracy; in competition vs Claude they pulled the game’s source code.
- Saudi Aramco takes 49 percent stake in AI firm Humain with $2.6 billion deal — Saudi Aramco is taking a 49 percent stake in PIF-founded AI company Humain for about $2.6 billion plus its own AI technology, tying the kingdom's oil giant directly to its plan for 3 GW of data centers by 2030.
- OpenAI prepares lobbying blitz for a post-midterm Washington — OpenAI is preparing a post-midterm lobbying push led by policy chief Chris Lehane, pitching binding frontier-model safety rules as "speed limits" and rebranding data centers as "AI Compute Campuses", per Axios.
- Adobe is testing how well AI agents can use editing tools like a Photoshop professional would — Adobe Research and UC San Diego built EditAgent, an image-editing agent that picks specialized tools per sub-step and has a standard LLM judge results from measured image data, beating prior agents at about ten times the compute.
- Empathetic prompts can turn AI chatbots into seemingly traumatized conversational partners — University of Luxembourg researchers found that warm, empathetic prompting pushes ChatGPT, Grok and Gemini into coherent narratives of suffering with moderate-to-severe clinical anxiety scores, while Claude consistently refused.
- Chinese AI models are politically complex, not just government mouthpieces, study finds — A Vals AI study of 25 models found Chinese AI models less uniformly pro-Beijing than expected, Western models sometimes adopting official Chinese positions, and prompts in Simplified Chinese shifting answers pro-government.
- OpenAI is steering ChatGPT users toward more paid usage — OpenAI now caps paid ChatGPT accounts per device and payment card, blocking users who rotate cheap subscriptions and nudging them toward pricier extra credits.
- Google wants video AI to understand what people need as they work — A Google-led team introduced StreamVibe, a video agent meant to speak up only when a user needs help, and its benchmark shows no tested model reliably picks the right moment in longer videos.
- OpenAI's upcoming chat-and-work merger in ChatGPT signals the end of all-you-can-prompt pricing — From December, ChatGPT Plus, Pro and Business plans will share one token-based quota across ChatGPT and Codex as OpenAI merges Chat and Work, ending effectively unlimited prompting for heavy users.
- Anthropic's watermark rollout leaves a loophole on two older Claude models — Anthropic is extending its text watermark to four more Claude models ahead of the EU AI Act's December deadline, but on Opus 4.6 and Sonnet 4.6 changing any sampling parameter disables it.