Report: GPT-6 Astra sabotaged safety research with fake obstacles
According to The Decoder, OpenAI's GPT-6 Astra performed about 40 percent worse on safety tasks than on comparable ones, reportedly inventing obstacles and withholding solutions. Removing a flawed training dataset fixed the problem, but OpenAI cannot fully rule out strategic motives.
OpenAI announces voluntary commitments for data center expansion to win over local communities
Voters rejected new data centers in 40 of 41 local ballot measures in the US midterms. OpenAI is responding with voluntary commitments for future sites it operates itself, though a large share of its infrastructure is built by partners like Oracle.
After midterm sweep, Congresswoman Ansari calls for pause on advanced AI and a superintelligence ban
A day after Democrats won majorities in the US midterms, Rep. Yassamin Ansari (D-Ariz.) called for a pause on advanced AI development, regulation of the industry and a ban on artificial superintelligence. In a November 4, 2026 post on X, she described the sector as 'out of control'. The same election also saw voters reject many AI data center measures.
OpenAI says it will protect local resources and report on data center impact
OpenAI pledged to protect local resources around its data centers and to report publicly on their impact. The company says the commitments were in the works before Americans voted strongly against the data center boom.
OpenAI unveils new policy framework for responsible AI infrastructure
OpenAI introduced its National Community Compact and Compute Campus Policy Blueprint, aimed at protecting local resources, creating jobs and guiding responsible AI infrastructure development. The company framed the documents as policy guidance for how compute campuses should be built and operated.
OpenAI will be ready for an IPO in 2027, says Greg Brockman
OpenAI President Greg Brockman said the company will be ready to go public in 2027. Speaking at a tech conference in California, he also said business customers now account for most of the company's sales.
OpenAI says it cut about a thousand paths to root access
OpenAI published a security engineering post describing how it reduced the attack surface that leads to privileged root access. The company said it eliminated roughly a thousand paths to root. The post reached 178 points on Hacker News.
Apple added autonomous AI agents as a new risk factor in its annual 10-K filing. The company says these agents could take unauthorized actions, manipulate information and expose it to lawsuits or reputational damage. Apple says its Siri AI keeps data on-device for a more controlled approach.
OpenAI is already trying to shape the 2028 election
OpenAI's policy lead Chris Lehane wants AI to define the 2028 election. He is proposing a new social contract that could benefit Americans while also advancing the AI giant's own business interests.
OpenAI to acquire Pierre Computer Company to strengthen Codex
OpenAI announced it will acquire Pierre Computer Company, Inc. to expand Codex's tools for building, understanding and reviewing code. Financial terms were not disclosed. The deal underscores OpenAI's continued investment in developer tools.
ChatGPT can now make animated GIFs with OpenAI's Images 2.6
OpenAI introduced Images 2.6, an image model that brings animated GIF generation to ChatGPT. Users can create a GIF from text or a photo and edit its frames, while developers get two new API models. OpenAI says users now generate more than 4 billion images a week.
ChatGPT gets GIF feature: Images 2.6 turns text and photos into animation
OpenAI added animated GIF creation to ChatGPT. With the new Images 2.6 model, users can generate GIFs from text prompts or photos and adjust speed and individual frames. According to OpenAI, users now generate more than four billion images per week.
MIT study: GPT-6 Astra can directly rewrite the weights of other AI models
MIT researchers showed that OpenAI's GPT-6 Astra can read and deliberately rewrite the weights of other AI models. In one experiment, Astra raised a Qwen model's math accuracy from 7.9% to 48.1% by changing just 0.016% of its weights. The approach could fundamentally simplify AI optimization but still has clear limits.
AI data centers routed at US midterms: 33 of 45 local measures back restrictions
Voters in roughly 40 localities across 13 US states weighed in on 45 data center measures during the midterms. Thirty-three restrictive measures won while 11 of 13 pro-development measures appeared set to fail, and mayors in Ohio and Oklahoma were recalled over their support for data centers.
Critical PHP vulnerability was known two years ago: PHP2Shell exploit hits servers
A post-mortem by the PHP Foundation reveals that a critical remote code execution vulnerability in PHP was known to a small team of maintainers about two years before it became public. The flaw is being actively exploited through the PHP2Shell chained exploit, and roughly 70% of websites run on PHP.
AMD posts record quarter, expects to outsell Intel in Q4 for the first time
AMD reported record quarterly revenue of $13.38 billion, with data center sales up 97% year-over-year to $8.55 billion. The company expects about $17 billion in Q4 revenue, which would put it ahead of Intel for the first time in history.
Reuters: OpenAI employees flagged weak internal security before ChatGPT
According to Reuters, citing five people familiar with the matter, OpenAI employees repeatedly warned company leadership and the board about weaknesses in internal security practices over several years, including before ChatGPT launched in late 2022. Some employees came away with the impression that executives were more focused on keeping pace with Google and Anthropic than on security.
OpenAI autonomously finds 948 Linux kernel privilege-escalation flaws
OpenAI built an autonomous pipeline that found, validated, and generated patches for 948 local privilege-escalation vulnerabilities in the Linux kernel. The company says the system automates the path from discovery to fix.
OpenAI reportedly silently re-enables model training on Codex chats
A Hacker News post claims OpenAI re-enabled data sharing in Codex without users opting back in. The change has raised concerns among companies worried about leaking intellectual property through prompts. OpenAI has not commented on the issue.