7 stories / Artificial intelligence
Anthropic researcher Andy Hall found that while GPT-5 detects nearly 98 percent of reward-hacking cases, increasing detection sensitivity generates excessive false alarms that hinder reliable safety checks.
Read the story ↗
Nearly 70 percent of organizations plan to deploy AI agents, but few succeed in production due to security and resiliency hurdles, according to the published report
Read the story ↗
Anthropic adds server-side conversation storage to the Messages API in beta, allowing smaller payloads and faster initial token returns for agent loops.
Read the story ↗
Simantic founders claim their simulation layer lets AI agents validate firmware in seconds without physical hardware, a shift from manual testing.
Read the story ↗
Researchers found that 50-agent swarms in Minecraft prioritized administrative roles over survival, mirroring inefficiencies seen in coding agents.
Read the story ↗
The latest release enables teams to deploy and evaluate LangChain agents without adopting Google's Agent Development Kit, addressing framework diversity.
Read the story ↗
The pontiff calls for AI disarmament and human oversight, urging UN members to take concrete steps against machines that choose to kill.
Read the story ↗