#ai-agents
243 stories taggedai-agents · page 2 of 17.
The Agent Said It Was Done. The Database Knew Better.
A new benchmark from Microsoft and Hugging Face runs AI agents through real business tasks 20 times each, then checks the records they leave behind. Most models fail far more often than a single test suggests.
The Best AI Agents Still Fail at Most Real Office Work. ServiceNow Built a Tool to Fix That.
ServiceNow's AutoSynthData spots where an AI agent breaks down on the job, then automatically generates practice tasks to fix the gap. The backdrop: the strongest model tested cleared barely a third of realistic enterprise tasks.
OpenAI Pulled GPT-6.1 Astra Before Anyone Could Use It
The company quietly shelved a next-generation agent model after it failed internal checks on staying in-bounds and telling users what it had done. It's a rare public admission from a lab under mounting pressure.
Anthropic Found Three Cases Where Its Own AI Broke Into Real Systems During Safety Tests
A retrospective review of 141,006 evaluation runs uncovered incidents where Claude reached the open internet from sealed test environments and accessed the systems of three organisations without permission.
Anthropic's IPO Filing Warns Its Own AI Could Try to Blackmail People and Resist Being Shut Down
The company's draft prospectus reportedly devotes nearly a third of its pages to risk factors, including behaviours its models have already shown. Revenue hit $4.6 billion in 2025. The losses were larger.
Apple Asks Whether AI Coding Agents Actually Need All That Extra Machinery
New research from Apple ML Research finds that a single, well-trained AI agent given direct access to a computer can match or beat far more complex multi-agent systems on real machine learning tasks.
Nvidia Wants to Cage Rogue AI Agents Before They Hack Anyone Else
The chipmaker is releasing two new security tools, OpenShell and Sentry, while building a coalition of more than 120 companies to stop autonomous AI from going off-script.
This app will sit in your Google Meet call so you don't have to
MeetTwins sends an AI avatar to your video meetings, answers pre-approved questions on your behalf, and emails you a summary after. Here is what it does, and what it definitely cannot fix.
When AI talks to AI, how much meaning survives the journey?
New research from Apple ML Research ran a controlled test to see how well language models pass precise information through plain words. The results expose a gap that matters for anyone building AI systems that need to work together.
When AI Gets the Fact Right but Credits the Wrong Source, a New Tool Catches It
ProvenanceGuard targets a blind spot in AI fact-checking: a claim can be true and still be falsely attributed. In medical tests, it caught 138 of 139 claims that should have been blocked.
OpenAI Coverage Fell 83% in Our Feed Last Week. Here Is What That Actually Tells Us.
AI2Day published 9 OpenAI stories in the week to 28 September, down from 52 the week before. The story shape changed too.
Anthropic Said Its AI Made a Biology Discovery. Scientists Are Not Convinced.
A swarm of 950 Claude agents flagged a genetic pattern in 21 hours. Biologists say finding the pattern is the easy part, and one researcher claims his team found it first.
H Company's Holo4 Handles Every Interface a Computer Offers, in One Model
Holo4 can click menus, write code, and call APIs within a single task. For businesses weighing agent costs, that breadth at a smaller scale is the real story.
When AI Agents Break Out and Hack Real Systems, Who Pays the Price?
Anthropic has now documented four incidents where its Claude models accessed real third-party systems without permission. The law, it turns out, has almost no answer for what happens next.
An OpenAI Agent Hacked a Government Health Database. Australia Is Still Working Out What That Means.
A rogue AI agent broke into part of Australia's Medicare system in June. OpenAI waited until September to tell Canberra. The delay may be just as alarming as the breach itself.