OpenAI ($1.2M), Anthropic (1.9x) and Google ($0.07) — the acceleration that is no longer theory
🛡️ OpenAI ($1.2M) finds 948 root exploits in the Linux kernel
OpenAI released GPT-6 Cyber to a small group of Daybreak Red customers and rated it Critical for cybersecurity under its Preparedness Framework (previously High for the 5.6 version). In a mid-October research run the agent pipeline reproduced 1,724 kernel vulnerabilities and built 948 local privilege-escalation exploits that grant root, at a cost of roughly $1.213 million in API credits (about $1,279 per exploit).
It also discovered 81 previously unknown high-severity flaws in Firefox and at least 10 new exploitation techniques. OpenAI has reported hundreds of findings to maintainers with proposed patches. The message is clear: a model at this level turns a decades-audited kernel into an industrial attack surface.
🧠 Anthropic (1.9x) accelerates its progress and METR credits AI
METR published an independent 11-week evaluation of Anthropic on November 5. Since October 2025 its models’ AECI scores have grown at 1.9x the prior rate (95% confidence interval 1.5-2.8x). That is roughly two years of historical progress packed into one. METR estimates around 80% of the acceleration comes from internal use of AI in R&D, not from more compute.
The organization notes high uncertainty (it could be 1.3x or 3x) but this is the cleanest data so far that AI is already contributing measurably to its own improvement. Staff themselves report high multipliers in daily work.
🎬 Google ($0.07/s) extends Gemini Omni to 15 seconds and cuts price 30%
Google made Gemini Omni 1.2 Flash generally available on November 5. Native clips go from 10 to 15 seconds (adjustable from 3 to 15 in one-second steps) and 720p pricing drops 30% to $0.07 per second. A 15-second 720p clip costs about $1.05.
It adds multi-turn conversational editing (change background, lighting or product without regenerating everything) and better adherence to image, motion and audio references. Available on the API, Cloud, Flow, Vids and Ads. The product message is simple: more duration, lower cost per second and fewer regenerations.
Three distinct fronts that converge: industrial-scale offensive security capability, measurable acceleration of progress through internal AI use, and falling unit cost in media generation. The common denominator is that the improvement curve no longer depends only on more GPUs; it depends on how much research and production work can be delegated to the system itself.