GitHub releases the layered defense design of the agent execution platform, and OpenAI releases the instruction hierarchy training data IH-Challenge and model. Responses to prompt injection were received from both infrastructure design and training axes.
OpenAI acquired AI security evaluation platform Promptfoo, and Microsoft announced that Anthropic's Claude Cowork would be integrated into Microsoft 365 Copilot. The structure of the enterprise AI market is starting to change.
Two approaches to achieve local isolated execution of AI coding agents. On macOS, Agent Safehouse uses OS-native sandbox-exec for kernel-level restrictions, and on Windows, Codex uses the VM-based Windows sandbox.
A summary of GPT-5.3 Instant’s hallucination reductions and safety regressions, GPT-5.4’s computer use, Tool Search, and 1M-token context, plus Saguaro’s 5× inference speedups.
AWS has made OpenAI API compatibility for the Bedrock Mantle distributed inference engine generally available, letting existing OpenAI SDK code run against open-weight models such as DeepSeek and Mistral.
Five new features for Copilot coding agent — model selection, self-review, security scanning, custom agents, and CLI integration — plus bidirectional Figma-Codex integration via MCP. Also covers Copilot CLI GA and comparison with Claude Code Figma integration.
Anthropic accused three Chinese AI companies of distilling Claude, and on the same day OpenAI retired SWE-bench Verified. Training fraud and evaluation flaws exposed simultaneously on February 23, 2026.
Techniques and defenses from the MINJA, InjecMEM, and ToxicSkills campaigns that poison AI agents’ memory files, and the fact that GPT-5.3-Codex achieved a 72% exploit success rate on EVMbench released by OpenAI and Paradigm. This article organizes how AI becomes both a target of attacks and a weapon for attackers.
Generalized the scripts from the practice and optimization articles into a reusable framework and published it on GitHub. A walkthrough of how to use it and the design philosophy.
