No Australian or New Zealand organisation features among the roughly 50 partners granted access to Project Glasswing, Anthropic's landmark AI-driven cybersecurity initiative. Now a rival model from OpenAI has entered the field, with a broader access pathway that may offer ANZ security professionals a more realistic route to frontier AI defensive tooling.

Announced on 7 April, Project Glasswing gives selected organisations access to Claude Mythos Preview, an unreleased frontier AI model. Anthropic says the model has already identified thousands of previously unknown zero-day vulnerabilities across every major operating system and web browser.

Project Glasswing takes its name from the Greta oto, a butterfly whose wings are almost entirely transparent.

Launch partners include Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorgan Chase, the Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks. Access has been extended to more than 40 additional organisations. All are US-domiciled or US-headquartered.

Anthropic has committed up to USD$100 million in model usage credits to Glasswing partners. A further USD$4 million in direct donations has been pledged to open-source security organisations, including the Apache Software Foundation and the Linux Foundation's Alpha-Omega and OpenSSF programmes.

Claude Mythos Preview is not a specialised security scanner. It is a general-purpose frontier model whose coding and reasoning capabilities have proven highly effective at identifying subtle flaws that have eluded human analysts and conventional tools, the company states.

Anthropic has said it will not release Mythos Preview to the public. The company cited the model's dual-use risk, noting that capabilities effective at finding flaws are equally effective at enabling attacks. Access is restricted to vetted partners with defensive mandates.

OpenAI Enters the Field with GPT-5.4-Cyber

One week after Anthropic's announcement, OpenAI on 14 April unveiled GPT-5.4-Cyber, a fine-tuned variant of its GPT-5.4 flagship model purpose-built for defensive cybersecurity work.

Like Mythos Preview, GPT-5.4-Cyber is not publicly available. OpenAI describes it as "cyber-permissive" - a version with lower refusal boundaries for legitimate security tasks. Initial access is limited to vetted security vendors, organisations, and researchers.

A key new capability is binary reverse engineering: the ability to analyse compiled software for malware, vulnerabilities, and security weaknesses without needing access to source code. This is particularly relevant for assessing legacy and third-party systems where source code is unavailable.

Access runs through OpenAI's Trusted Access for Cyber (TAC) programme, which was launched in February 2026 alongside a USD$10 million cybersecurity grant programme. TAC is now being significantly expanded.

How ANZ Organisations Can Apply for Access

OpenAI's access model differs significantly from Anthropic's closed partner coalition. Rather than invitation-only participation, TAC uses tiered identity verification, with higher levels of verification unlocking more powerful capabilities.

Users approved for the highest verification tier gain access to GPT-5.4-Cyber, which carries fewer restrictions on sensitive tasks such as vulnerability research and analysis. According to SiliconANGLE, OpenAI is targeting thousands of individual defenders and hundreds of security teams.

The contrasting approaches were noted by Constellation Research, which observed that OpenAI appeared to take a deliberate jab at Anthropic's partner-selection model, arguing that identity-based access is preferable to ecosystem partnerships for ensuring broad defensive coverage.