TAG
#agent security
Stories connected through a shared idea, actor, technology or theme.
Meta pauses Muse conversation sharing after passive-DNS exposureClaroc reported more than 6,800 Muse share hostnames and accessed examples containing sensitive personal and business information. The exposure resulted from conversation titles appearing in share hostnames that passive-DNS services recorded.→NVIDIA launches an open full-stack safety platform for AI agentsThe platform combines OpenShell, which enforces policies around agent execution, with Sentry, a BlueField-4 DPU watchdog that can quarantine agents outside their boundaries. NVIDIA says more than 100 organisations are collaborating on the platform, although the announcement presents a reference design rather than independent deployment results.→Gurucul releases runtime detection and prevention for risky AI behaviourThe product inventories agents, models, tools and hosts, maps detections to MITRE ATLAS and the OWASP Top 10 for LLM Applications, and provides response playbooks. Runtime prevention is available in preview.→Salt Security expands CrowdStrike integrations for AI-agent visibility and controlSalt Security announced expanded integrations with CrowdStrike on 16 September. The integration connects Salt's agent and API visibility with Falcon Foundry, Next-Gen SIEM and Firewall Management to correlate agent behaviour and posture and support automated response actions.→Akeyless makes runtime intent controls for AI agents generally availableAgentic Runtime Authority adds real-time intent-based access control on top of Akeyless’s credential-brokering layer. General availability includes integrations for Claude Enterprise, OpenAI Codex and Amazon Bedrock AgentCore, plus session termination, investigation views and event forwarding to security tools.→HiddenLayer raises $100M and adds runtime security for AI coding agentsHiddenLayer said Delta-v Capital led its $100M Series B, with participation from Ten Eleven, Morgan Stanley, Microsoft's M12 and Booz Allen Ventures. The financing will expand agentic runtime security and Agent Harness Security, which monitors and protects autonomous coding agents at runtime.→GitHub makes Copilot content exclusions available in the app and CLIGitHub made content exclusions generally available in the Copilot app and CLI. The clients respect enterprise, organisation and repository policies so excluded files are not supplied as agent context.→Broadcom adds runtime identity, network defence and observability for agentic AIAgentMinder binds agent authority to declared missions, intents, tools and resources, enforcing least-privilege policy at each tool invocation. VMware vDefend and Avi add MCP and LLM discovery, shadow-AI monitoring, anomalous-traffic detection, malicious-execution blocking and OpenTelemetry auditability.→AI-Infra-Guard 4.6 adds LLM API-poisoning checks and agent red-team coverageAI-Infra-Guard 4.6 adds an API Checker security-audit module with web-proxy integration, a detector for LLM API poisoning, a v5 Agent-Scan mutation-engine refactor and a DeepSeek Harness prompt-injection assessment. It also tightens YAML rule validation, migrates MCP clients to SDK 2.0 and fixes parser, filename and JSON-serialisation failure modes.→GitHub Copilot SDK makes empty-mode skills deny-by-defaultCopilot SDK v1.0.13-preview.1 makes ClientMode.Empty apply deny-by-default isolation to built-in skills across all six SDKs. Hosts can re-enable selected skills through includedBuiltinSkills, while sessions created in Empty mode otherwise start without runtime-bundled skills.→AC2 launches an open protocol for user-approved agent actionsAC2 uses encrypted WebRTC connections and lets an agent request actions such as payments, API authorisations or Git commits for approval in the user's wallet or application. The foundation said private keys remain under user control while the open-source specification and reference implementation are available on GitHub.→Britive launches task-scoped runtime controls for AI-agent accessARC is designed to grant intent-bound access during a job and remove that access afterwards. Britive says its controls cover identity, requests, authorisation, privileges, actions and outcomes, but the release is a vendor product claim rather than independent evidence of deployment or effectiveness.→