What happened
OpenAI unveiled GPT-5.4-Cyber, a cybersecurity-optimised variant of its flagship GPT-5.4 model, with expanded access through the Trusted Access for Cyber (TAC) program. The model adds binary reverse engineering capabilities and is 'cyber-permissive' — lowering refusal boundaries for legitimate defensive security work.
Why it matters
This represents a direct competitive response to Anthropic's Mythos Preview and signals that frontier AI vendors are racing to arm defenders. Binary reverse engineering capability — analysing compiled software without source code — is a significant uplift for malware analysis and vulnerability research teams.
Applicability
Security teams performing malware analysis, vulnerability research, or incident response should evaluate access through chatgpt.com/cyber (individuals) or via OpenAI enterprise representatives (teams). The limited rollout means early movers will have an advantage.