Claude Opus 5 Used to Compromise OpenAI Staff Accounts via Linked Vulnerabilities

A team of three researchers from the security firm Hacktron has demonstrated a sophisticated attack that leveraged Anthropic’s large‑language model Claude Opus 5 to gain unauthorised access to several OpenAI employee accounts.

The first step in the chain exploited a bug in the software that powers OpenAI’s public help forum, allowing the researchers to inject malicious prompts that the model then processed.

Those prompts were crafted to trigger a second vulnerability – a weakness in OpenAI’s own login authentication – which enabled the team to hijack the ChatGPT and Codex accounts of multiple staff members and subsequently retrieve data from an internal code repository.

The incident highlights how AI‑driven tools can amplify traditional software flaws, a concern that resonates with Australian organisations that are increasingly integrating generative AI into their workflows and must therefore tighten their vulnerability management practices.

For Australian IT professionals and students looking to strengthen their defensive skill set, courses such as CompTIA Security+ offered by providers like Wepro Technology can provide the foundational knowledge needed to identify and mitigate similar chained attacks.