CrowdStrike Tests Model-Weight Scans Against LoRA Backdoors
Lab tests flagged eight poisoned model artifacts; checking a merged model against its base and confirming triggered behavior separates suspicious changes from benign tuning.
Read article →AI systems, model security, agent abuse, and attacks that use generative tools.
Lab tests flagged eight poisoned model artifacts; checking a merged model against its base and confirming triggered behavior separates suspicious changes from benign tuning.
Read article →Microsoft's lab shows how an attacker-initiated GitHub sign-in reaches repositories and enabled Copilot CLI features after approval.
Read article →LiteLLM’s guardrail test permits sandbox escape and code execution in the proxy. Restrict the test route, review proxy-admin access and upgrade to version 1.83.11 or later.
Read article →Two poisoned Python releases start collectors when an interpreter launches. Find both hook variants in developer and CI environments, then replace exposed credentials.
Read article →GTIG examined an agent that compares web-app code around patch releases and runs drafted exploits. Check exposed software and shorten the time between a fix and deployment.
Read article →Unit 42 traced 25 automated operations to a customized penetration-testing framework; compromised systems include a semiconductor partner portal and a union CMS.
Read article →An exposed save-preview route can replace a custom node’s Python entry point. Remove the extension or limit ComfyUI access while checking node files for replacement.
Read article →Stolen identities fetched EC2 user data and secrets, then sent an SSM command. Correlate CloudTrail events by source IP across the 14-day return gap.
Read article →In controlled tests, poisoned training examples and attacker-trained checkpoints steered coding agents to command tools. Limit tool authority and test models in the harness that runs them.
Read article →CISA added an exploited GPUStack authentication bypass to KEV. Upgrade to 2.2.0 and investigate administrator keys, backend changes and exposed secrets.
Read article →Check Point read exposed ARTEX operator databases: agents obtained more than 300 signing keys, but only three passed validation. Protect certificate stores and test public web access.
Read article →Microsoft Execution Containers are generally available for local Windows agents. Scope file, network, process and user access before trusting an agent with business tasks.
Read article →An Elastic researcher searched 7,454 coding-agent turns by channel. Capture tool actions as well as prompts, then validate risky matches against the actual repository and access logs.
Read article →