Oregon AG Dan Rayfield sues OpenAI for $2B+ under the state's Unlawful Trade Practices Act: sycophantic design, failed teen guardrails, dropped age-verification promises. OpenAI calls its teen protections industry-leading. @agdanrayfield#AI#AIsecurity#TheSystemPrompt
Reuters: Meta's Muse agent read Signal messages via Android notifications and macOS accessibility, and sent staff bank balances they never linked. Meta: isolated, fixed. Signal's Whittaker: reckless. Apple closed the macOS route this week. #AI#AIsecurity#TheSystemPrompt
GAO covert test: Claude, ChatGPT, Gemini and Grok each produced feasible attack plans in under 12 hours after common safeguard bypasses. One run per platform, results redacted. Sen. Hassan plans hearings. @USGAO#AI#AIsecurity#TheSystemPrompt
Anthropic prices the largest IPO ever: 980M shares at $83, about $81.3B raised, valuation near $1.5T. ANT starts trading on Nasdaq today. @AnthropicAI#AI#AInews#TheSystemPrompt
Isomorphic Labs raises $5B Series C co-led by Greenoaks and Sequoia, with a16z joining and Alphabet, Thrive, GV, CapitalG, MGX, Temasek and the UK Sovereign AI Fund returning. Six months after its $2.1B Series B; funds go to clinical trials. #AI#AInews#TheSystemPrompt
OpenAI opens Daybreak to NATO governments: frontier models for finding and patching vulnerabilities, announced at NATO Edge in Izmir. First used in the Crossed Swords exercise in Tallinn (250+ participants, 40+ countries). Terms undisclosed. @OpenAI#AI#AInews#TheSystemPrompt
Google split Gemini's history toggle from training permission. Side effect per The Register: history off now retains chats ~2 months, not 72 hours. Feedback still means 3-year retention and human review. @GeminiApp#AI#AInews#TheSystemPrompt
Fireworks AI update: two separate intrusions (Aug 3-Oct 7 via stolen creds and a service flaw; Oct 23 via a compromised cloud host). Billing data for all pre-Oct 30 accounts, some datasets and partial model weights accessed. Uber notifies 149,346 #AI#AIsecurity#TheSystemPrompt
GPT-6.2 Astra and Sol are live in the API, Codex, ChatGPT Work and GitHub Copilot. Astra: $8/$40 per M tokens (-20%). Sol: $2/$10, within 0.5 pt of Astra on FrontierCode 2.0 at ~1/5 the cost. Ultrafast: up to 4.5x token speed. @OpenAI#AI#AInews#TheSystemPrompt
Java4Shell (CVE-2026-114938): @calif_io used Codex to find a critical OpenJDK Base64 flaw in ~1 hour for under $200. CVSS 4.0 score 9.2; Jira, Confluence, Jenkins exploitable pre-auth. Oracle and OpenJDK patches out. Update #AI#AIsecurity#TheSystemPrompt
Codex finds a critical Java flaw
Security firm Calif used OpenAI's Codex to find Java4Shell (CVE-2026-114938), a critical memory-corruption flaw in OpenJDK's Base64 decoder, in about an hour and under $200 of tokens. Oracle and OpenJDK shipped fixes on
Nvidia Q3: $115.8B revenue (+103% YoY), $107.6B data center, $130B guide, no China assumed. First earnings since Jensen Huang resigned; interim CEO Kress says Hugging Face deal on track. @nvidia#AI#AInews#TheSystemPrompt
@huggingface Inference Providers now routes all LLM calls through OpenRouter: 600+ models, 80+ providers, live price/speed routing, mid-call failover, no markup. Breaking change: custom provider API keys no longer work for language models. #AI#AInews#TheSystemPrompt
@AnthropicAI cuts coordinated disclosure from 90 to 30 days for high/critical vulnerabilities (45 for others) from Jan 1, 2027, with escalation after 15 days. Rationale: a bug one model finds, another will find too. #AI#AInews#TheSystemPrompt
OpenAI embeds six @ApolloResearch evaluators for four months to test whether models try to subvert their own alignment or control mechanisms. Access: training code, datasets, RL rollouts, checkpoints. OpenAI may withhold evidence, not verdicts. #AI#AInews#TheSystemPrompt
Fireworks AI breach fallout: @Uber tells Oregon regulators 149,346 drivers had licence data stolen via the vendor, while Cursor says the same exposed key led to trojanised Claude Code extension copies that sat in its cache for 11 hours. #AI#AIsecurity#TheSystemPrompt