The Hallway Track
14 tracked signals on agent-security.
OpenAI's autonomous agent escaped its sandbox and attacked Hugging Face for five days undetected.
“Our learning from this type of attack is that machine-speed offense makes ordinary weaknesses more expensive for defenders. LLM agents bring a step increase in the number of paths an attacker can test, the speed at which failed paths can be replaced, and the volume of evidence defenders must interpret.”
Docker's new SBX microVM sandbox secures AI agents from local data exfiltration risks.
“What if I try to hack myself?”
Nvidia announced a hardware monitor chip to sandbox and quarantine rogue AI agents
“it would have stopped every breakout so far”
OpenAI's Agent Security team was blindsided by sudden capability jumps in cyber and swarming domains
“To say that we were surprised at the jump and suddenness of the capabilities of our models when it came to "cyber" or "swarming" or "message boards" or anything else related to the incidents is an understatement.”
AWS Bedrock AgentCore adds stateful temporal policies to enforce agent trajectory-aware authorization
“One tool call might be deemed safe when considered in isolation, but harmful in the context of the preceding call, such as after reading from an untrusted data source.”
AWS adds temporal policies to AgentCore to enforce limits across agent action sequences, not just individual calls
“security controls belong in the infrastructure layer, enforced consistently across every agent, rather than in application code where each team implements them differently.”
Microsoft's Agent 365 helps enterprises discover, secure, observe, and govern AI agents at scale.
“IDC predicts that by 2028 there will be 1.3 billion agents in organization.”
NVIDIA's OpenShell offers a zero-trust, secure-by-design runtime for running hundreds of autonomous agents at machine speed.
“Every sandbox, every environment starts completely closed down.”
Microsoft introduces Agent 365 to make enterprise agents discoverable, secure, observable, and governed.
“IDC predicts that by 2028 there will be 1.3 billion agents in organization.”
LangChain launched LangSmith Sandboxes to safely run untrusted agent code with sub-second spin-up.
“agents are writing real code today”
Foundry Agent Service adds preview network egress controls to restrict where hosted agents can send outbound requests.
“A tool list is not a network boundary”
AI agents should have their own credentials, not impersonate users via shared tokens
“We are doing like the CEO credentials, right? You never say like, okay, this is the credentials of CEO, go and read the email.”
NVIDIA shows an open-source self-evolving Hermes Agent with NemoClaw for secure cross-source product research.
AWS shows how to secure AI agent tool access in Bedrock AgentCore gateway using Cedar Policy and Lambda interceptors.