AI6 min read
DiffuPlan: A Search Agent with an Editable Planning State
The DiffuPlan search agent revises its plan as new clues arrive, reaching an average F1 of 50.24 across four multi-hop QA benchmarks.
AI news, AI tools, automation
Blog
AI news, new models and integrations, AI tools, automation and practical notes. Every article carries a date and, where there are any, its sources.
AI2 min read
ECHOTRIBBLE: US Army Plans Autonomous AI Cyberweapons
Internal documents show Army Cyber Command is building AI agents that move through foreign networks like a worm. Experts warn of collateral damage.
AI2 min read
AI Agents Alone on a Virtual Island: Do They Play?
Copies of Claude Opus 4.7 invented their own games on a simulated island. Researchers see playful learning; skeptics urge caution.
AI2 min read
AI agents break into more than 100 websites
A hacker used autonomous AI agents to attack more than 100 websites, says Norforge. Victims reportedly include well-known industrial firms.
AI2 min read
AI Agents on the Web: Who Benefits From the New Web?
Personal AI agents have arrived on the web. Many sites block them, others open up. That could change the web for humans.
AI2 min read
AI Agents Pay via Stripe Link: First Noticeable Revenue
Upstarts Media reports that AI agents increasingly shop for users via Stripe's payment tool Link. What's behind it and what matters.
AI2 min read
Kling AI 4.0: Longer Clips and More Control
Kling AI 4.0 offers clips up to 30 seconds, ten keyframes and 8,000 tokens of prompt length. It doesn't lead against Wan 3.0 and Seedance 2.5.
AI2 min read
Vetting LLM Judges Without Reference Data: A Case Study
An arXiv study presents a staged pipeline that pre-screens LLM judges and models before expert review, using a dental hygiene coach as the example.
AI5 min read
Microsoft: AI Agents as an Insider Risk in Companies
Microsoft advises companies to closely monitor AI agents that employees use with systems outside the big AI labs.
AI6 min read
OR-License: How Good Are LLM Agents at Operations Research?
New OR-License benchmark tests whether LLM agents can turn real client documents into usable optimization solutions. No model reaches expert level.
AI6 min read
Anthropic gives METR employee-like access
Anthropic opens its systems to nonprofit METR: external reviewers assess risks of Claude models with access similar to that of an employee.
AI2 min read
Anthropic Has Nearly Stopped Hiring Junior Staff
At Anthropic, the junior share of new hires fell from 27 to 9 percent, while AI agents work six times as many hours as humans.