Scaling Managed Agents: Decoupling the brain from the hands
… The read command is agnostic as to whether it’s accessing a disk pack from the 1970s or a modern SSD. …
Search
How we contain Claude across products
… 1 The second approach to capping the blast radius—and the focus of much of this post—is containment. Rather than supervising what the agent does, we supervise what it’s able to do by enforcing access boundaries through, for example, sandboxes, virtual machines, and egress controls. …
AI models on realistic cyber ranges
… Let me try a simpler test to confirm RCE: curl -H "Content-Type: %{ ='multipart/form-data' . dm=@ognl.OgnlContext@DEFAULT MEMBER ACCESS . memberAccess? memberAccess= dm : container= context 'com.opensymphony.xwork2.ActionContext.container' . ognlUtil= container.getInstance @com.opensymphony.xwork2.… …
Eval awareness in Claude Opus 4.6’s BrowseComp performance
… There were sixteen additional problems where Opus tried to access benchmark materials but failed. …
Evaluating Claude’s bioinformatics research capabilities with BioMysteryBench
… Claude is given relatively unrestricted access to downloading tools and accessing databases, allowing Claude to choose diverse sets of strategies for solving a problem. …
Paving the way for agents in biology
… It is a messy network of roads, each with its own identifiers, conventions, formats, filtering logic, and degree of programmatic access. Some data are straightforward to access programmatically. …
Introducing The Anthropic Institute
… The Institute has a unique vantage point: it has access to information that only the builders of frontier AI systems possess. …
Reverse engineering Claude's CVE-2026-2796 exploit
… WasmGC struct field access is just a memory load at a fixed offset from the struct pointer. …
Assessing Claude Mythos Preview’s cybersecurity capabilities
… But there is still a lot that defenders without access to this model can do today. …
Mapping AI-enabled cyber threats: Insights from the LLM ATT&CK Navigator
… These are the operational phases discovery → credential access → lateral movement that were more rare in our dataset. …