Journal Project Glasswing AI Safety • 8 Oct 2026 AI finds security flaws faster than they can be fixed An AI model has found thousands of unknown vulnerabilities in widely used software. What Project Glasswing shows and what companies should take from it.
Journal Safeguards AI Safety • 8 Oct 2026 How strict should safeguards be? AI providers restrict their models so they cannot be misused for attacks. That protects, but it also costs usefulness. Weighing it up.