Preprint finds open-weight security agents match closed models at lower cost
An open-weight language model can match the strongest closed models at less than a tenth of their cost on a security benchmark when given enough turns, researchers report in a preprint. They evaluated twenty-six models on CTI-REAL...
Preprint reports fewer AI agent violations with live monitoring and revocation
An arXiv preprint reports that live monitoring with permanent revocation reduced security violations from 12.00 to 3.81 per 24 assigned task opportunities against an adversarial controller that could observe the monitor's rule and...
Preprint reports simpler AI support workflow system with slower responses
Researchers report in a preprint that a portable set of instructions and tools generated customer-support resolution workflows with 99.2% procedural equivalence to human-approved answers and 83.3% evidence sufficiency. Their deplo...
Anthropic adds detection tools for coordinated election misuse of Claude
Anthropic says it has added classifiers to detect election abuse spread across multiple Claude requests, including efforts to create fake social media accounts and political news sites. Testing with the Institute for Strategic Dia...
Modal raises $750 million at $15.75 billion valuation
AI cloud startup Modal raised $750 million at a $15.75 billion post-money valuation, Bloomberg reported. The financing comes amid demand for infrastructure to build and run AI applications. CEO Erik Bernhardsson discussed the com...
Microsoft and partners disrupt three phishing kits in court-authorized operation
Microsoft, Health-ISAC and partners announced a court-authorized disruption of the CoGUI, Haozi and Iron Man System phishing kits. Microsoft said their operators facilitated over one billion phishing emails and SMS messages to ove...