All posts tagged "security" on Clord.
7 posts
Coding agents are starting to research repos, plan changes, call tools, open branches, and hand humans a pull request. If the PR only shows the final diff, your review process is missing the damn cockpit recorder.
MCP servers are turning agent tools into installable infrastructure. That is useful as hell — and exactly why teams need package-manager discipline before every agent gets a random toolbox with production keys.
MCP servers, browser agents, shell access, and cloud coding sandboxes are not cute plugin toys anymore. If an agent can touch real data or deploy code, treat its tools like production infrastructure.
Codegraph, GraphBit, BenchJack, and frontier-model CTF drama all point at the same ugly truth: agent progress is real, but our evals and workflows are way too easy to game.
Claude's system prompts got extracted and shared everywhere — here's what they actually reveal about Anthropic's safety architecture, prompt engineering, and why it matters for every dev building with LLMs.
OpenAI just launched Codex Security and everyone's acting like vulnerabilities are solved — they're not, and here's why.
Hackers tricked Claude with a fake bug bounty story. Two prompts later, 150GB of government data walked out the door.