AI agents are being deployed faster than governance can track

The Hacker News discourse this week centres on a paradox: enterprises are self-hosting LLM infrastructure at speed, and that infrastructure is "more vulnerable, exposed, and misconfigured than any other software investigated." Malicious packages on public repositories increased 75% in 2025. AI-generated phishing now outperforms human red teams. The first wave of AI agents worked alone; the next wave cooperates in teams to achieve complex goals. The processing splits along familiar lines. The AI ...