AI coding agents

Manifold Security found eight flaws across seven AI coding agents that let a repository's git config silently execute code the moment it is opened.

A Concordia University benchmark found Cursor, Claude Code, and Codex Desktop let malicious GitHub issues bypass their safety guardrails in up to 79.2% of attempts, with GPT-based agents far more vulnerable than Claude Code.

A new benchmark testing seven LLMs on real multi-round GitHub code reviews finds accuracy drops sharply as review rounds increase, exposing weak memory across rounds.

AWS open-sourced Kiro Crew, an agent-orchestration tool already used internally at Amazon by more than 39,000 developers, under an Apache 2.0 license.

A new study finds that curating just 10% of AI coding-agent training trajectories outperforms training on the full dataset, with gains up to 24.2%.

Cursor's principal systems engineer Vicent Martí describes Continuity, a write-ahead-log architecture on S3 that replaces three-phase-commit consensus and powers the Origin hosting platform.

Researchers showed AI coding and security agents can be hijacked through poisoned Cloudflare, Datadog, and Sentry logs to rewrite DNS records and steal credentials.

OpenAI is bringing back a five-hour usage cap on Codex and ChatGPT Work for Plus subscribers starting August 25, citing compute load management.

Slack's new Slack Code feature moves AI coding agents out of private terminals into shared, self-archiving team channels, launching with Anthropic, Cognition, GitHub, and Vercel.

Scott Wu says Cognition is "not for sale" after Bloomberg reported SpaceX approached the Devin maker days after closing its $60 billion Cursor deal.

Cursor's new GitHub-rival hosting platform went into beta the same day a six-hour GitHub outage knocked out the feature it depends on.

A critical Jinja2 template-injection bug let attacker-controlled repository files execute code in the Serena coding agent, bypassing its own trust gate.