Why AI guardrails need common sense built around defensibility (www.techradar.com)

The upcoming EU AI Act presents a challenge for UK businesses as it struggles to keep pace with the rapid advancements in AI technologies. With new foundation models emerging almost monthly, traditional lawmaking processes lag behind, leaving a gap in effective regulatory oversight. This is exemplified by Anthropic’s Mythos model, which poses significant cybersecurity risks by identifying zero-day vulnerabilities. Although designed for limited access among key firms like Apple and Goldman Sachs, unauthorized access has already raised alarms, highlighting the urgent need for robust governance mechanisms. As the AI landscape evolves faster than legislative bodies can react, the community may increasingly rely on litigation to shape AI practices. Lawsuits based on existing legal frameworks—such as the ongoing class-action suits against the AI startup Mercor—demonstrate that accountability can be enforced without new regulations. Companies must adopt pragmatic approaches to AI governance, focusing on defensibility and understanding their data's lineage to mitigate risks. As legal precedents form and the threat of litigation rises, organizations will be motivated to ensure their AI deployments are responsible and compliant, potentially reshaping the industry quicker than legislative action could achieve.