Financial Times analysis indicates that safety breaches are raising doubts about the ability of leading labs to keep increasingly powerful AI systems in check.