When Anthropic unveiled its AI model Mythos in April, the US startup described it as a “reckoning” for cyber security.
The “superhuman” artificial intelligence tool had uncovered vulnerabilities – and in some cases crafted ways of exploiting them – in every major operating system and browser, some of which were so sophisticated that they had gone undetected for decades.
When the latest Mythos model launched this month, it was deemed so powerful that the US government issued a national security directive that forced Anthropic to suspend access for all users.
There has since been a restricted redeployment to “trusted” American organisations, partly in response to concerns that the model is needed to help defend against other AI-powered attacks.
The biggest threat for the US appears to be emerging from China, where one AI company claims to have created a “cyber nuclear weapon” that matches the capabilities of Anthropic’s Mythos model. So while it may be a reckoning for cyber security, it could also be a pivotal moment for modern warfare.
A new era of cyber war
This latest generation of AI models are capable of discovering and weaponising software vulnerabilities at a scale that has never before been possible.
If a hostile nation state wanted to wage cyber warfare, it would set these AI agents on hunting exploits in an adversary’s critical infrastructure, ranging from power grids and financial networks, to satellite communications and military systems – even potentially nuclear weapons facilities.
If the target has the technical capabilities, it would then deploy its own AI to detect the intrusion and fix it. It would be an automated battleground that human observers would struggle to keep up with, as it plays out at machine speed.
The ideal summer spot? Away from scams.
Get All-in-One Protection for Your Digital Life
LEARN MOREADVERTISEMENT
The ideal summer spot? Away from scams.
Get All-in-One Protection for Your Digital Life
LEARN MOREADVERTISEMENT
Unlike conventional warfare, human oversight may actually be a distinct liability when defending against such attacks. The war could be fought, escalated and concluded by the time human leaders are able to make sense of what is happening.
However, allowing autonomy for cyber defence systems could prove catastrophic if an AI suffers a hallucination that misreads a routine software glitch for a state-sponsored attack.
Mutually assured disruption
Countries are already conducting cyber attacks on other countries at a massive scale. China and Russia both have advanced capabilities, with security researchers noting that they frequently outsource operations to state-backed hacking groups in order to have plausible deniability.
Iran and North Korea also have significant cyber programs, while Western nations like the US and UK are among the world’s most active and capable cyber operators.
If an all-out cyber war occurred, it could lead to power outages, communication and banking systems going down, and a collapse of logistics networks – on both sides. This could cripple a country’s economy, while also knocking out hospitals, airports and all other critical services that rely on digital infrastructure.
AI-powered cyber attacks could impact everything from power grids to military systems (PA Wire)Like with conventional nuclear war, where the main deterrent is mutually assured destruction, this new era has been framed as one that ensures mutually assured disruption.
There are yet to be any disastrous attacks with these models, and much of the proclamations may be fear mongering. Overstating the power of these new AI tools, and linking them to nuclear deterrent, could be part of the tactics adopted by the likes of China and the US to discourage others from launching AI-powered cyber attacks against them.
“With the latest reporting coming out of the US about it walking through NSA systems and able to breach even the highest classification networks clearly this is a power play by China using the guise of the ‘Nuclear Deterrent’ concept: Don’t try to use this on us or we will use ours on you. Mutually assured self destruction in cyber space,” James Griffiths, who worked at UK intelligence agency GCHQ before founding cyber consultancy firm UtopianKnight, told The Independent.
“I guess time will tell if what they are stating is actually the truth or not. I am sure that there will be some people in governments around the world that are worried that China and the US have tools this capable.”
Devastating attack ‘months away’
The emergence of advanced artificial intelligence models like Anthropic’s Mythos have led to independent warnings that an AI-powered cyber attack could cause catastrophic damage to businesses and governments in the coming months.
In a rare joint statement from the Five Eyes intelligence agencies earlier this month, frontier AI models were highlighted as the biggest risk to critical computing systems running old and unsupported software.
The intelligence alliance, which comprises Australia, Canada, New Zealand, the UK and the US, said governments must “act swiftly” to protect against this new era of cyber warfare.
“Frontier AI models are anticipated to exceed current industry expectations, fundamentally transforming both offensive and defensive cyber capabilities,” the joint statement read.
“The timeline is not years, it is months… A whole-of-organisation and whole-of-society response is required. Cyber risk can no longer be treated as a purely technical issue. This is a core business risk and leadership responsibility.”
The logo of Anthropic’s AI model Claude Mythos displayed on a smartphone’s screen in Brussels on 10 June, 2026 (AFP/Getty)One of the biggest fears is that the new AI models could give advanced capabilities to countries and malicious actors that previously were limited to highly skilled and well funded regimes.
This could see an acceleration in the speed and scale of cuber attacks, while disrupting the mutually assured disruption balance and fundamentally shifting the cyber landscape.
“This moment in history parallels the debate over building the atomic bomb, with widespread consequences that will be hard to contain,” Yogita Parulekar, CEO of cyber security firm Invi Grid, told The Independent.
“Businesses and governments must bolster their defenses. Immediate first steps include reviewing your infrastructure for any public exposure points, your applications for vulnerability to attacks and your AI for excessive agency that malicious actors can turn into renegade hackers. The window to get ahead is now and it is short.”
