Anthropic has billed Claude Mythos Preview as “a substantial leap” in terms of cybersecurity, with the San Francisco-based company reporting in April that the model had already discovered “thousands of […] high- and critical-severity vulnerabilities.”
Yet even after holding several meetings with EU officials in recent weeks, there have been no direct talks on giving European entities access to Claude Mythos, raising strong cybersecurity fears for the continent.
However, OpenAI has announced it will be providing the EU with access to its latest ChatGPT 5.5 model, which the UK’s AI Security Institute (AISI) has recently tested, confirming that it “reaches a similar level of performance” on cyber tasks.
Alarm bells in Europe
Officials have been busy raising alarm bells about Mythos since Anthropic revealed it in early April, with the ECB’s Frank Elderson urging European banks this week to prepare quickly for a wave of attacks launched with the assistance of the new model.
Elderson also suggested that the EU’s lack of access to Mythos made the situation even worse, heightening the need to act quickly.
“Lack of access is not an excuse for inaction,” he said. “On the contrary, it makes it even more critical that banks step up and act now.”
While the EU still doesn’t have access to Mythos, it does have access to the next best thing, ChatGPT 5.5, which available benchmarks suggest is at least as capable as its rival.
“Lack of access is not an excuse for inaction”
European Central Bank board member Frank Elderson
According to the AISI, ChatGPT 5.5 is the second model (after Mythos) to have successfully completed one of its multi-step cyber-attack simulations, from start to finish.
This suggests that GPT 5.5 represents the same kind of stepchange as Mythos, yet the AI Security Institute found that OpenAI’s model performs better than Anthropic’s on certain key metrics.
Most notably, ChatGPT 5.5 scored a pass rate of 71.4% on the AISI’s advanced suite of tasks, at the expert (as opposed to practitioner) level.
By contrast, Claude Mythos scored 68.6% at the same level, leading the AISI — which is a directorate of the UK’s Department of Science, Innovation, and Technology — to conclude that on this particular measure, “GPT-5.5 may be the strongest model we have tested.”
OpenAI for the EU
Given that OpenAI has now agreed to provide the EU with access to ChatGPT 5.5, the AISI’s report would indicate that European institutions can now protect themselves against cyberattacks as well as they could with Claude Mythos.
Some of the European companies to have already accessed GPT 5.5 include Deutsche Telekom, BBVA, Telefonica, Sophos, and Scalable Capital, with dozens of other companies included.
‘GPT-5.5 may be the strongest model we have tested’
AI Security Institute
EU access has been granted via the developer’s OpenAI for Countries initiative, which is led by former UK treasury minister George Osborne, who joined the company in December.
Osborne said in a letter published on Monday, “Through the OpenAI EU Cyber Action Plan, we will work with European policymakers, institutions and businesses by democratizing access to the defensive tools that trusted actors can use to strengthen shared security, support public safety and reflect European priorities.”
The appointment of Osborne, who for five years was the second-most powerful politician in the UK, is an interesting one, since his background in politics would suggest that the OpenAI for Countries initiative is as much about government lobbying as anything else.
According to OpenAI’s chief global affairs officer, Chris Lehane, Osborne’s addition to the company reflects a belief that early decisions about how AI is “built, governed, and deployed will shape economics and geopolitics for years to come.”
OpenAI’s offer to the EU could therefore be seen as part of efforts to improve relations with the EU, which is currently evaluating whether to classify ChatGPT as a Very Large Online Search Engine.
Anthropic has billed Claude Mythos Preview as “a substantial leap” in terms of cybersecurity, with the San Francisco-based company reporting in April that the model had already discovered “thousands of [...] high- and critical-severity vulnerabilities.”
Yet even after holding several meetings with EU officials in recent weeks, there have been no direct talks on giving European entities access to Claude Mythos, raising strong cybersecurity fears for the continent.
However, OpenAI has announced it will be providing the EU with access to its latest ChatGPT 5.5 model, which the UK’s AI Security Institute (AISI) has recently tested, confirming that it “reaches a similar level of performance” on cyber tasks.
