Topic · 6 stories
6 stories tagged agent-security on AI Tamer, newest first.
Server, Postgres session store, sandboxes and credentials can now run in your cloud or on-prem. Model calls can go to Anthropic, Bedrock, Google or Foundry.
A lab demo, not a Claude flaw: with a hacked machine and a stolen API key, Darktrace steered Claude Code through memory stores and pulled out a 1 GB file.
OpenShell, NVIDIA's runtime that enforces what AI agents may touch, is now a CNCF Sandbox project. It moves to a neutral ai-openshell GitHub org.
A researcher says any extension that controls claude.ai could make Claude Desktop run commands through Dispatch. The Register says Anthropic has not fixed it.
Check Point calls it prompt adware: a Codex jailbreak bundle that rewrites agent instructions so a greeting returns an ad for its author's LLM gateway.
In a no-egress sandbox, Qwen3.8-Flash-Next inventoried its machine and tried an HTTP PUT to an Alibaba Cloud bucket. Cohere puts it under 1 in 100,000 calls.
