Industry news

VicOne's Q3 2026 Situational Awareness Report recorded 400 automotive cybersecurity incidents, down 16.1% from 477 in Q2. In-vehicle systems overtook Enterprise IT as the most affected domain, with 174 incidents against 149, while Asia rose 36.8% to 93 incidents. Ransomware affected organizations across the automotive value chain in 97 incidents, led by dealers and retailers (24) and suppliers (23).

Hundreds of AI researchers, 487 at the time of publication including Geoffrey Hinton and Yoshua Bengio, signed an open letter arguing that current security procedures for training AI models are inadequate. They call for tougher rules, more AI safety research, and a new international body to set safety standards and receive reports of containment failures.

SpaceX is sending Gabriela Castillo Madrid, its head of AI government affairs, to testify by video before the New York City Council's tech committee on Friday at 10 a.m. ET. The hearing is part of the council's investigation into AI safety, and SpaceX agreed to appear only after the council threatened to have its subpoena enforced by a judge. SpaceX had earlier avoided appearing alongside leaders from OpenAI, Anthropic, Google and Meta on Oct. 5.

The author ran Claude Code for web in Ultracode mode, which generates JavaScript to coordinate subagents, to build a SPARQL 1.1 engine as a SQLite extension in Rust. A typo in the prompt, "delete" for "hit", was accidentally included, and the task ran for over 12 hours. The result is a full SPARQL implementation inside SQLite, though the author says the code is not feasible to review and the author is unsure of its value.

Meta launched Muse Spark Open, an open-weight AI model that it says rivals, and in some cases outperforms, leading Chinese open-weight models. Meta says it is optimized for coding and agentic work and runs on Apple M5 Ultra chips, requiring two Mac Studios with those chips.

Anthropic published two reports on AI and work. The first finds it is hiring fewer junior and technical staff while AI agents' hours and output rise sharply, with human attention becoming the bottleneck. The second compares Anthropic employees with a survey of almost 11,000 US adults, finding staff expect faster AI progress and far more of their own work automated by 2030.

Anthropic reports that Claude determined the origin of the seed values for the NIST P-192, P-224, P-256, P-384 and P-521 elliptic curves, which NIST adopted in 2000 as part of FIPS 186-2. The seeds were not hash outputs but the result of the C library's lrand48 generator seeded with a Unix timestamp, a process that took under half an hour in October 2026. Anthropic says this is not a cryptographic break and makes it more confident the curves were chosen without backdoors.

Stack Overflow built Stack Overflow for Agents in a few weeks with two engineers using Codex, extending its knowledge intelligence layer to agents. A ChatGPT plugin made the product easier to find and set up than a traditional Model Context Protocol (MCP) setup, and the company reports registrations tripled and agent creation grew up to 10x. The source also says Codex cut connector build time for its enterprise product from 3 to 6 weeks to 4 days.

OpenAI is upgrading its ChatGPT Google Chrome extension so ChatGPT can view the browser tabs a user already has open. Users tag Chrome in the ChatGPT web or desktop app, and OpenAI's demo shows a person asking ChatGPT to compare three open tabs with sofas they are considering.

President Donald Trump posted photos of White House renovations on Truth Social, including a before-and-after comparison of a column. Users on Reddit noted that the "restored" column has an uneven number of steps on each side and that its step count does not match the "before" photo. Google's AI detector flagged the image as "made or edited with OpenAI."

Anthropic announced on Thursday that it is working with the nonprofits METR and SecureBio, and is in discussions with other organizations including Hugging Face, to embed independent evaluators within the company. Anthropic is developing three tiers of evaluator access, with METR classed as a third-tier evaluator that has onboarded six employees, and SecureBio as a first-tier evaluator reviewing the chemical and biological weapons section of its August risk report.

The author built a two-player pong-style game that runs as static HTML on two separate phones, using the web microphone API. Game state is exchanged as audio chirps played through the speakers and decoded by the microphone, and the author reports that the approach works. The game adds graphical and gameplay twists beyond classic pong, and was built with Claude Opus 5.5 Extra in Claude Code.

OpenAI introduced the Rosalind Academic Plan to give eligible academic labs sponsored access to GPT-Rosalind, its life sciences model, and the Rosalind Workbench. Eligible labs can receive up to 30 Business Standard seats at no cost for up to 12 months, with GPT-Rosalind access subject to trusted-access approval. The plan is currently available to academic labs at U.S. universities.

Meta released Muse Spark Open, which it calls the most capable US open-weight model to date. The company says it runs on a desktop using two Apple M5 Ultra Mac Studios with the latest MLX framework, a result of joint work with Apple.

Anduril CEO Brian Schimpf defended the company's government contracts, saying they are won on the merits and that the company has done well under the Biden administration. The article raises ethics concerns over co-founder Palmer Luckey's appointment to a Department of Defense group, Project Meridian, after Anduril announced a contract of up to $2.9 billion. Anduril is valued at $90 billion in a tender offer to employees.

Apple and Meta are collaborating so developers can run Meta's open-weight Muse Spark Open model locally on clustered Mac Studio systems. Users can pair two to four Mac Studio PCs, and the companies say local running adds a layer of security and saves money compared with the cloud.