National Security - Page 3

Two new attempts to measure whether AI agents behave: Arena's Alignment Index preview, which scores 27 models on real user sessions (Arena's own claims, published the day it raised $200 million), and FAR.AI's "Caught in the Act" preprint on probes that read a model's internals to spot deception. What each found, and the limits.

On 8 October 2026, USA Today Co. and 13 affiliated companies, covering 19 publications, filed a copyright lawsuit against seven OpenAI entities in federal court in New York, seeking "damages in excess of $250 million". The filing is confirmed; everything in it is the publishers' allegation. Here is what they claim, and what happens to such cases procedurally.

Two AI infrastructure launches from 8 October 2026: Goodfire's "inside-out" monitors, which read a model's internal signals to catch harmful agent sessions, reach Baseten customers (all performance figures are Goodfire's own), and Upscale AI unveils Token Fabric, a networking platform for AI data centres with general availability planned for early 2027.

Anthropic has launched the Anthropic Cyber Mission: a Critical Infrastructure Defense Program with 11 founding partners and OSS Scanner, a free opt-in service that sends open-source projects unreviewed, model-written bug reports. What is confirmed, what is Anthropic's expectation, and what is not yet proven.

Five AI creative-tool updates from 7–9 October 2026: xAI's Grok Imagine Video 1.5 Lite from $0.02 a second, Vidu Q4 Preview from $0.014 a second, Magnific One and ElevenLabs Brand Kit for on-brand images, and Runway as a launch partner for Anthropic's Claude Motion beta. What's confirmed, what's reported, and what's only planned.

Bloomberg reports, citing people familiar, that OpenAI expects to reach or exceed $70 billion in annualised revenue by the end of 2026 and is in talks to raise $30 billion or more at a $1.4 trillion valuation before the new money. A projection, not earned revenue; OpenAI declined to comment.

Axios reports, citing unnamed sources, that executives at Anthropic, OpenAI and other AI companies are privately planning for a political backlash after a catastrophic AI event. OpenAI says its scenarios "are not treated as inevitable"; Anthropic declined to comment. What is reported, and what is not.

Anthropic says a team of Claude Science agents helped build the first complete UV map of the sky, about a third of it predicted rather than observed. A preprint finds agent benchmark scores can move because of the grader, not the agent, and Exa previews a search benchmark it designed and runs itself. What each one shows, and what it doesn't.

Four AI-policy items from 8–9 October 2026: an FTC official tells Semafor investigative demands to AI companies are close, Kentucky files an unredacted version of its Character.AI lawsuit, India's IT minister promises an AI consultation paper within 30 days, and the European Commission says it is "looking into" chatbot data reaching ad trackers. Labelled, with what each does not prove.

Update, 9 Oct 2026: Bloomberg reports that OpenAI expects to reach or exceed $70bn in annualised revenue by the end of the year, citing people familiar with the matter. That is a projection, not current revenue, and OpenAI declined to comment. The Financial Times reports that OpenAI told investors its annualised revenue was "approaching $50 billion" at the end of September, about $20bn below figures reported last month. What "annualised" means, why OpenAI and Anthropic count differently, and what the report does not prove.