Researchers identify 627 malicious ClawHub skills using instruction and code analysis
Researchers report identifying 627 malicious skills among 66,157 ClawHub skills after screening with a new security analysis framework and confirming findings through auxiliary dynamic validation and manual review. Their preprint ...
Researchers report agent defense that reduces harm without increasing missed emergencies
Researchers report that a trust-gating defense produced the lowest measured harm on every model tested in SentryBench, a new benchmark for AI agents using public-safety tools. The preprint evaluates attacks alongside failures to r...
Anthropic filing identifies gas supply as a constraint on data center plans
Anthropic says some planned data centers depend on electricity from on-site natural gas turbines, exposing its compute expansion to fuel and pipeline constraints, The Verge reported from the company's Monday securities filing. Wit...
Meta opens AI network transport software and submits switch platform specification
Meta says it is making its MetaRoCE network transport software available as open source, starting with a user-space implementation on GitHub called libsoftmetaroce. The transport addresses communication between accelerators in AI ...
SWI Compute orders about $1bn of Nvidia systems for AI infrastructure
SWI Group's new computing business has ordered about $1bn of Nvidia systems and plans a UK AI campus that would start at 30MW and expand to 530MW, the company said in a Monday press release. The initial UK facility is due online ...
Anthropic filing reveals first operating profit and own data center effort
Anthropic reported its first operating profit, $974 million in the second quarter, in the IPO prospectus it filed Monday, Yahoo Finance reported. Revenue rose to $11.6 billion from $4.7 billion in the first quarter, while the cost...
