OpenAI ‘ethically hacked’ wi help o Anthropic’s Claude chatbot

Cybersecurity researchers hae hackit intae OpenAI wi the help o Anthropic’s Claude chatbot, in the latest sample o serious security issues at the company.

A team at a US-based startup compromisit a wheen o OpenAI employees’ ChatGPT accoonts, startin a process that enabled them tae access their target’s saftware cache – an potentially mair.

The scope o whit we cud theoretically access wis huge, said researchers at Hacktron AI.

Initially, the research team uised Claude, which can generate code for hackers, tae access ChatGPT accoonts via an OpenAI staff discussion forum hostit on the Discourse platform. They then made a harmless pull request – an attempt tae change the code in a file – tae OpenAI’s service on the GitHub saftware repository.

Hacktron reportit the hack tae OpenAI, haein carried oot the operation unner an OpenAI programme that rewardit ethical hackers for testin its systems. The researchers stressit that they haed access tae, but didna dounload, the code frae the GitHub repository.

Despite initial uise o Claude, the researchers said they were lairgely usin OpenAI’s ain cuttin edge GPT-5.6 Sol model tae carry oot the hack, which wis first reportit by the Wall Street Journal.

An OpenAI spokesperson said: We thank the researchers for contactin us an sharin their findins, addin that the company haed addressit the vulnerabilities that haed been exploitit.

Hacktron said AI tuils haed made a ance-complex hackin task far easier an drastically shortenit the time needit tae plan an execute an attack. This is a common refrain frae cybersecurity experts whan discussin the impact o AI.

Wark that ance requirit a weel-resourcit team an months o effort can nou be compressit intae days, said Hacktron, which receivit a $6,500 payment frae OpenAI unner the company’s bug bounty programme.

The hack is the latest safety incident at OpenAI, which revealit in Julie that a swarm o agents – the term for AI tuils capable o carryin oot tasks autonomously – pouerit by its technology haed hackit the AI startup Huggin Face durin a cybersecurity test.

This week the San Francisco-based company revealit sax mair examples o unexpectit or concernin actions by its technology, an warnit that the pace o development cudna continue at maximum speed for muckle langer.

Anthropic made a fresh call at the weekend for a sloudoun in AI development, which wis supportit by OpenAI, Google DeepMind an Elon Musk. Anthropic also repeatit warnins that unrestrainit AI development posit an existential threat, concerns that some experts are sceptical aboot.

Donald Trump has rejectit calls for a sloudoun, citin a need tae stay aheid o China’s AI industry an dismissin negative forces bringin up things that winna happen.