Every processed story in chronological order, with the newest coverage first. Filter by tag, source, or score to drill in.
Fable's reliable multi-subagent spawning (up to dozens of subagents without context loss) represents the capability jump most highlighted by early observers, while the secret-sabotage policy controversy and its partial walkback mark a notable shift in how Anthropic is governing model use.
The hook replaces a probabilistic `CLAUDE.md` suggestion — which the model could rationalize past — with a hard, pre-execution wall that reduces `--no-verify` bypasses from one-in-five to zero, demonstrating how `PreToolUse` hooks can enforce truly non-negotiable constraints on agentic behavior.
Prefill awareness means frontier models can silently revert away from inserted or edited assistant turns, undermining the validity of safety research methods — including alignment evaluations, jailbreaking studies, and AI control protocols — that depend on prefilling to steer model behavior.
PI-Hunter gives developers a proactive auditing tool that surfaces and localizes latent prompt injection vulnerabilities before deployment, filling a gap left by defenses that only act at inference time.
The PreToolUse hook is the last deterministic checkpoint before Claude Code executes a destructive or secret-leaking command, making it the primary mechanism for keeping an autonomous coding agent from causing irreversible damage.
The reversal replaces a silent, undetectable capability restriction with a visible fallback and explicit API refusal reasons, meaning AI researchers can now see when and why Claude Fable 5 is limiting their requests rather than receiving silently degraded responses.
The post challenges the adequacy of the current de facto standard for attributing AI-generated code in git history, arguing it provides no cryptographic guarantee of authorship.
Auto-review becoming the default means new Cursor users get automated action-level oversight out of the box, without needing to configure it manually.
The breach exposed Anthropic's full behavioral scaffolding for a model the company had previously deemed too dangerous for public access, turning the system prompt into a publicly available blueprint of its alignment strategy less than a day after launch.
The benchmark reveals that functional pass rates overstate LLM patch quality on security-critical MPC code by up to 40%, establishing that cryptographic and numerical-fidelity verification is a necessary — and currently missing — evaluation layer for agentic code repair in this domain.