Post
1 Nov 2026
ChatGPT will end chats fixated on violence, OpenAI says
OpenAI says ChatGPT will be able to automatically end chats when someone is fixating on violence, and may temporarily pause accounts after multiple such endings. For all users it will give less detail about past mass attacks, including weapons used; teens get tighter limits on practical weapons information even without signs of violent intent. OpenAI made over 200 law enforcement referrals in Q3 2026 and will publish its first Community Safety Report in early 2027. Rollout begins in the next few weeks.
02CodingChatGPT and Codex will share one usage limit in December
OpenAI's Help Center says that starting in December, Plus, Pro and Business plans' usage allowance will be shared across ChatGPT and Codex as it brings the agentic capabilities of Work into the main ChatGPT experience. Usage will be measured in tokens; more advanced models and complex tasks generally use more, and Instant chat will not draw from usage. OpenAI expects most users won't need to wait for resets more often than now; combined usage appears in Settings → Usage, and at the limit you can wait, upgrade, or buy credits.
OpenAI Help Center · help.openai.com ↗ (opens in a new tab)
03AgentsDeepSeek dsh-superagent: open-source coding agent
DeepSeek open-sourced dsh-superagent, a proactive coding agent for one GitHub repository packaged for DeepSeek Harness. It listens to GitHub and optionally Feishu; when an allowed person asks, it opens a work item, works on it in its own agent session and directory, and answers in the original conversation, for example with a pull request. Its rule is to take as little human attention as possible: it acts only when asked or when a team-written rule allows it. DeepSeek calls it a research prototype, not an official product; MIT licensed.
GitHub (deepseek-ai) · github.com ↗ (opens in a new tab)
04AgentsNVIDIA AI agents now steer a SETI telescope
NVIDIA says Breakthrough Listen and the SETI Institute have brought agentic AI to the Allen Telescope Array in Northern California: from one natural language request, agents can steer the instrument, correct incoming data and run AI models on the live stream. Built on NVIDIA Holoscan, the system handles 2.4 terabits per second with inference on site, and NVIDIA says agents compress weeks or years of work into hours. The pipeline made what NVIDIA calls the first real-time AI detection of a pulsar, in the Crab Nebula.
NVIDIA Blog · blogs.nvidia.com ↗ (opens in a new tab)
05SecurityChatGPT bug let one shared link take over accounts
Researchers at Parameter Labs showed that a malicious public ChatGPT share link could run attacker-controlled JavaScript inside the signed-in session of whoever opened it. In their test it listed the victim's conversations and retrieved content from a private conversation; opening the link was the only required action. The attack chain started with two harmless features: a graph, whose math-keyboard listener accepted messages from the sandboxed HTML Preview, and the HTML Preview itself. OpenAI fixed it on October 26, rated the report P1 and awarded a $5,000 bug bounty.
Parameter Labs · parameter.ai ↗ (opens in a new tab)
06ModelsGrok API adds invisible text watermark from Nov 7
From November 7, 2026, text from seven Grok API models (grok-4.7, 4.6, 4.5, 4.3 and the grok-4.20 family) carries an invisible statistical watermark by default, which the docs say is to comply with the EU AI Act. It shapes token choice rather than adding characters, and covers ordinary text, structured outputs and tool call arguments. Detection is statistical and access to it is by application; team admins can opt out per model only after confirming a legal exception applies.
Grok API docs · docs.x.ai ↗ (opens in a new tab)
07AI NewsPlanAssist AI planning tool expands to 15 English councils
Google DeepMind's PlanAssist, built with the UK government, Google Cloud and Faculty, moves from pilots in Camden, Barnet and Dorset to 15 English councils for a 12-month Beta on live applications. In the pilots it cut average officer time on householder assessment tasks by 58%, and officers approved over 70% of its generated reports with no edits. It checks documents, heritage constraints and planning policy and drafts cited reports; planning officers still make every decision. The government intends to offer it to every council in England by the end of 2027, subject to successful testing.
GOV.UK · gov.uk ↗ (opens in a new tab)
08AI NewsGoogle, OpenAI, Microsoft back one AI training opt-out
On Oct 30, 2026 Google, OpenAI and Microsoft committed to honor one shared robots.txt policy token, __usage_ai-training, so websites can state a preference against AI training once instead of writing separate rules for each provider's crawler. Google plans to treat an opt-out as a Google-Extended disallow and aims for 2027; Microsoft starts February 1, 2027 and says Bing search eligibility is unchanged; OpenAI says the coming months. The proposal is an individual IETF Internet-Draft written by Anthropic's Kevin Kelley; it is voluntary and does not block crawling.
Google Search Central Blog · developers.google.com ↗ (opens in a new tab)