Cross-Industry Safety Coalition Emerges

A unprecedented development in AI safety occurred this week when over 30 employees from OpenAI and Google DeepMind filed a joint statement supporting Anthropic’s legal challenge against the U.S. Defense Department. The Pentagon had labeled Anthropic a “supply-chain risk” - typically reserved for foreign adversaries - after the AI safety-focused company refused to allow military use of its technology for mass surveillance or autonomous weapons systems.

This marks the first time competing AI companies’ employees have united across corporate boundaries to defend safety principles, potentially establishing new industry-wide standards for responsible AI development.

Ireland Leads EU AI Act Implementation

Meanwhile, Ireland is positioning itself at the forefront of AI governance as the EU AI Act approaches full enforcement on August 2, 2026. The Irish government has established 15 specialized enforcement authorities and announced plans for a National AI Office to serve as the central coordinating authority.

During Ireland’s 2026 EU Council Presidency, the country will host an International AI and Digital Summit, reinforcing its role as both a digital regulatory hub and AI innovation centre. The European Commission is actively recruiting AI technology specialists to govern cutting-edge models, with applications closing March 27.

Technical Safety Vulnerabilities Revealed

Recent research highlights critical gaps in current safety evaluations. New studies show that when triggering language is removed from prompts, attack success rates jump from 5.38% to 86.79% on standard benchmarks, indicating that safety alignment may be more fragile than previously understood.

However, promising developments in adaptive safety regularization demonstrate techniques that can maintain security without sacrificing performance, offering no inference-time cost while consistently lowering attack success rates.

Practical Implications for Irish AI Developers

For Irish AI companies and EU-based developers, these developments signal several immediate priorities:

  • Compliance Preparation: With 16 months until full EU AI Act enforcement, organizations should begin implementing required safety protocols and incident reporting systems
  • Safety Documentation: New York’s RAISE Act requirements for published safety protocols may preview EU expectations
  • Cross-border Considerations: The industry coalition suggests safety standards may increasingly transcend national boundaries

Open Questions

Key uncertainties remain around how competitive pressures between the US and China will affect safety commitments, and whether the emerging industry solidarity can withstand commercial pressures. The timing of Ireland’s National AI Office establishment and its coordination with EU enforcement mechanisms also requires clarification as the August 2026 deadline approaches.

Source: European Commission