Published on 6/25/2026
The electronic intelligence agencies of the Five Eyes Alliance countries warned in a rare joint statement of the dangers of the development of artificial intelligence technologies and the approaching spread of models with advanced cyber capabilities that could accelerate cyber attacks against governments and companies. The statement made clear that these models are months away, not years, as was thought, according to a report published by the British newspaper The Guardian last Monday, June 22.
The three-page statement stressed the need for urgent action to confront the threat posed by artificial intelligence models, stressing that it is no longer possible to deal with cyber risks as a purely technical issue, but rather as a fundamental commercial risk and leadership responsibility that threatens the peace and stability of countries.
Read also
list of 2 itemsend of list
In a related context, the Reuters news agency report published at the same time explained that the intelligence agencies’ statement lacked details, in addition to its emphasis on basic cybersecurity advice, such as quickly repairing defective software and not connecting systems to the Internet except when necessary.
However, the statement urged defense officials to use artificial intelligence to strengthen cyber defenses, by identifying vulnerabilities early and responding faster to incidents, stressing that cyber resilience in this high-risk environment is an integral part of the mechanisms for confronting threats.
This sudden public intervention by the intelligence alliance that includes Australia, the United States, New Zealand, Canada, and the United Kingdom comes against the backdrop of the decision of the administration of United States President Donald Trump to ban foreign nationals’ access to the “Fable” model of artificial intelligence that was previously presented by Anthropic.
It is assumed that the “Fable 5” model is a more suitable version for public use and use than the more advanced model that “Anthropic” presented under the name “Methos” and boasted of its cyber capabilities, after it initially refused to launch the model due to the risks it carries.
The “Methos” model can discover security vulnerabilities in various systems, then develop malicious software that exploits these vulnerabilities without any significant interference from human hackers.
The researchers, who obtained prior access to the model several months ago, were able to exploit its capabilities to discover and fix security vulnerabilities in the famous Firefox browser, as well as penetrate Apple’s defenses.
Mythos is not alone
Although the American decision singled out Anthropic’s “Mythos” and “Fable 5” models, these models are not alone in the cybersecurity arena, as OpenAI developed a model carrying advanced cyber capabilities under the name “GPT 5.5 Cyber”, and it was not part of the American ban.
The statement expects that these advanced models will begin to emerge from all companies working on developing artificial intelligence models, as a company’s access to them means that others will be able to access them as well.
For her part, Olivia Shen, an expert in national security and artificial intelligence at the Center for United States Studies at the University of Sydney, explained the possibility that there are other models that mimic “Mithos” and are being developed by countries such as China or other parties, but we cannot see them and do not know anything about them.
“We should expect that the next Mythos or Fable model is coming very soon,” she added, which is directly consistent with the Five Eyes statement.
Is “Mythos” worth all the hype?
A report by the American technical website “Spaceworks” published on April 22, following the initial disclosure of “Methos” and documents related to it, questions the capabilities of the model and the extent of the threat it poses to the cybersecurity sector.
The report also explained that “Anthropic” had exaggerated in describing the size of the vulnerabilities that “Methos” was able to discover in the statement announcing the model and the “Glasswing” project responsible for it, because the statement estimated the number of vulnerabilities it discovered in the thousands, but the research documents available for review about the model clearly stated that the number of vulnerabilities reviewed by the human team behind “Methos” was only 198 vulnerabilities.
What Anthropic did was to take the human reviewers’ agreement rate with the model’s initial vulnerabilities that were reviewed, and then generalize this rate to all the vulnerabilities found by the model, which is a logical fallacy that assumes that human reviewers will agree on a fixed rate when reviewing all bugs.
Despite this, the SpaceWorks report does not marginalize the capabilities of “Methos” as an advanced artificial intelligence model, as the model has high logical thinking capabilities that make it stronger than previous generations of artificial intelligence models.
It should also be noted that the Five Eyes Statement does not refer to the “Mythos” or “Anthropic” model directly, but generalizes the warning against AI models in general.
