AI Engineering Signal #152
OpenAI agents found two critical PHP vulnerabilities in core PHP
Signals
OpenAI agents found two critical PHP vulnerabilities in core PHP
patch all hosts to PHP 8.2.35 immediately; audit WAF rules for new exploit signatures.
Latent Space
Compromised LiteLLM 1.107.0 on PyPI steals credentials
Audit CI lockfiles, rotate secrets exposed during the compromise window immediately.
Web
eMudhra intermediate CA keys cracked in milliseconds
Distrust eMudhra intermediates; reissue affected certificates and monitor TLS for forged chains.
Web
Recursive launches Flint, a 40B laptop coding model built largely by AI
Validate AI-built model outputs before deployment; local 40B coding agents now compete with cloud APIs.
TechCrunch
Agent swarm analyzed millions of mouse brain cells in 32 minutes
Add multi-agent biological screening with human expert adjudication gates to discovery pipelines.
Web
SimHack detects reward hacking onset before RL training
Run pre-training hack-risk gates before launching RL runs to avoid policy exploits.
ArXiv
The Take
Supply-chain integrity and emergency patching are now the binding constraints on AI infrastructure, while local models and agent swarms are moving into production fast enough to require expert gates before deployment.
Subscribe
Related Signals