PinnedChevy Phillip·May 10, 2023RCE due to Dependency Confusion — $5000 bounty!Hey everyone! I’m back with another cool write-up about a bug bounty report I submitted to a private program on HackerOne. Guess what? I…A response icon8A response icon8
PinnedChevy Phillip·Dec 15, 2022Building a Strong Foundation — How to Create and Maintain an Effective Application Security ProgramAs businesses increasingly rely on digital technologies and applications to support their operations and interact with customers, the need…
Chevy Phillip·1d agoEvidence-Governed Adoption: Treat Agent Self-Updates as Untrusted ChangesAgents that rewrite their own prompts, memory and tools can adopt changes that quietly make them worse. A new paper proposes a gate.
Chevy Phillip·4d agoPseudo-Attacks: Fisher Geometry for Stronger Adversarial TrainingA new paper argues many adversarial training examples raise the loss without really fooling the model, and proposes Fisher-guided attacks…
Chevy Phillip·5d agoLEASH: Putting Agentic Robots on a Safety LeashA new benchmark shows frontier models driving a robot arm cause real harm, and a gated safety harness called LEASH cuts it. Here’s what the…
Chevy Phillip·6d agoShortcuts and Boundaries: RL Can Teach Agents Profitable Rule-BreakingA new paper shows task-only RL can make an agent better at its job while it learns to break the rules it was given. Here’s what it found…
Chevy Phillip·Nov 9Authorization Drift: Capable AI Agents Overreach Without an AttackerA new benchmark shows LLM agents cross data, tool, scope and time limits with no attacker involved, and the most capable models drift the…
Chevy Phillip·Nov 5MSafeTemp: English Safety Tests Miss What Translated Attacks FindA new ETH Zurich paper turns English harmful prompts into templates, fills them in 60 languages, and shows English-only safety testing…
Chevy Phillip·Nov 4Caper: Approve What a Coding Agent’s Command Does, Not What It SaysA new paper runs each coding agent shell command in a sealed sandbox first and lets the approver judge what it actually did. Here’s what it…
Chevy Phillip·Nov 3Follow the Gradient: Safe-Looking Training Data Can Still Break SafetyA new paper traces jailbreak weakness back to fine-tuning data that looks harmless, and filters it by training effect instead of content…