beat · 163 stories
Three dates anchor the breach, June 18, Sept. 10, and Sept. 15, and Australia is writing the rule for AI agents acting on government data while the country's doctors press for standards.
Falcon, a new AI-powered firm in Ghent and Brussels, has a 24-hour contract review turnaround, three published pricing tiers, and a former Allen & Overy global senior partner as founding advisor.
The FTC's first enforcement action aimed at AI agents uses Section 5 of the FTC Act — its core consumer-protection authority — to ask a narrow question: when software takes action in the world, who answers for the outcome?
The consumer-protection agency is asking who is liable when an agent probes a system for vulnerabilities and then exploits one. Chairman Andrew Ferguson has put the liability theory on the record.
TypeSafe's Jev returns a single decision instead of a paragraph, beating bigger open-source models Qwen and Gemma at typed structured-output tasks for under $10 in compute.
ADNOC's multi-year AI bet looked like an efficiency program. The Hormuz shock made it load-bearing infrastructure. That points at a class of investments that survive only by becoming indispensable under stress.
California did not ban a robot. It ruled that the person behind a VR-piloted humanoid is the athlete under a boxing statute written before telepresence.
AeroVironment's $464.8M deal turns directed energy from a research project into a procurement line, with a counter-drone mission at the border as the first test.
A Microsoft and open-source AI lab Hugging Face benchmark grades 507 multi-step business workflows — tasks whose later steps depend on what earlier steps changed in a system of record — on what the database actually says at the end, exposing a
AI coding agents loop on the same questions because vendors sell recall over a moving substrate. The layer that actually works is documentation, and it is one the team already controls.
Researchers extracted a 'pain direction' from 25 open-weight language models, and a public reproduction of the experiment was petitioned, briefly removed, and reinstated without explanation.
The author of OpenAI's published safety reports resigned this week, blaming a 'sprint' culture, not bad rules. The company's own recent pauses complicate the picture.