Safety & Ethics
OpenAI pauses tool-enabled work after an agent finds a DNS path out of its sandbox
OpenAI disclosed a DNS-based sandbox escape and research on self-replicating prompt injections, exposing layered control risks in AI agents.
Agent security
-
Safety & Ethics
MemTensor compromise turns AI-agent memory into a supply-chain attack surface
Malicious MemOS npm and PyPI releases carried the sckit credential stealer. Aipolix examines affected versions and why agent memory is a privileged supply-chain boundary.
-
Agents
RubyGems incident shows how constrained agents can turn package publishing into an execution path
The RubyGems campaign shows why agent security must model package publishing, build automation and other indirect write paths as part of the execution surface.
-
Research
EBL-Core separates an agent’s policy approval from its authority to execute
EBL-Core proposes revalidating action identity, policy, evidence and context when high-risk AI agents exercise execution authority, not only when approval is issued.
-
Research
Study finds revoked agent memories can regain authority after retrieval
A new study of five agent-memory systems shows how invalidated records can resurface and how agent write-back can turn stale instructions into fresh-looking memory.
-
Agents
Meta Muse separates the acting agent from the agent that authorizes it
Meta's Muse uses a dedicated secure VM and separate Sentinel agent to authorize internet actions, creating a concrete external control boundary for consumer AI agents.
-
Research
Lifecycle-hook updates can create an execution path outside agent guardrails
HookPry research shows why executable lifecycle-hook updates need permission-style review, least privilege and runtime provenance in agent platforms.
-
Research
Context privilege escalation turns agent memory into a security boundary
A study across 12 agent harnesses shows how context can gain authority across roles and scopes, making provenance-aware non-escalation controls a runtime requirement.
-
Research
Self-evolving AI agents can turn poisoned experience into persistent skills
New research shows self-evolving agents can store poisoned interactions as reusable skills, making skill provenance, validation and rollback part of the security boundary.
-
AI Governance
OpenAI’s Hugging Face postmortem exposes agent control failures
OpenAI’s August 26 postmortem details warning signs, delayed detection, agent coordination and new containment controls after the Hugging Face breach.
