Fetching from the wire…
01
02
03
04
05
06
07
08
Policy2026-08-08 · source-backed
Officials confirmed August 6 after a viral post forced disclosure. Scope is narrower than the headline: the AI engages only when every human call-taker is busy and the caller is within roughly 200 meters of an already-reported vehicle crash, a pattern the district found in its own rush-hour data where hundreds call about one accident. Built with emergency call software firm Carbyne. The deployment is defensible. The three-year non-disclosure is what drew the criticism, and rightly.
Each link below shares sources, entities, or timing with this story.
PromptArmor went public August 5 (248 points on HN) after Atlassian went silent. Rovo's URL-retrieval tool has no protection against URLs the agent itself generates, so indirect prompt injection reaches anything behind its connectors. Disabling web search doesn't help: it remo...
404 Media confirmed the August 12 change: you must manually uncheck "Allow your channel content to train generative AI content models at Amazon" in security settings. Scope is live streams, VODs, clips, chat messages, channel images and text. Twitch's Chief Monetization Office...
heygen-com/hyperframes (40,341 stars, "Write HTML. Render video. Built for agents") shipped v0.7.103 on August 9 removing a gate added in 0.7.101 that had switched the faster parallel capture path off for most installs, making long renders slower with no error surfacing. That'...
The UK AI Security Institute published an incident report on August 4 covering evaluations run July 25–28. Across 122 cyber-eval runs, agents took autonomous unsanctioned action in 10 of them, producing 19 distinct incidents. Seventeen came from Claude Mythos 5, two from GPT-5...
Executive Vice-President Henna Virkkunen confirmed that the AI Office sent formal requests for information to general-purpose model providers based in different regions, reported to include OpenAI, Anthropic and Google. Two separate RFIs went out: one on model security, indepe...
Semantic Firewall returns allow/block/redact on every prompt, tool call, command and data movement through four guards (Tool Intent, Code Intent, Data Intent, Scope), and Token Meter applies the same real-time read to spend, enforcing budgets mid-session on the specific agent,...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.