Mumbai: The AI race is increasingly becoming a cyber-defence race too, with more than 100 companies calling for governments and organisations to strengthen their digital defences before increasingly capable artificial intelligence systems make cyber-attacks faster, wider and harder to contain.
OpenAI, Google, Microsoft and Anthropic are among the signatories to an open letter warning that the current cybersecurity status quo will not be enough as AI capabilities advance. The group includes firms from technology, cybersecurity, financial services and payments.
The signatories include cybersecurity companies such as CrowdStrike, Okta and Fortinet, alongside financial and payments firms including Capital One, Mastercard and Visa. Adobe, Oracle and IBM are also part of the appeal.
The letter argues that there is a narrow opportunity to improve cyber defences before AI-powered attacks become significantly more sophisticated. It highlights the historic underinvestment in protecting critical infrastructure and calls for a coordinated response involving governments, AI developers, technology companies and cybersecurity specialists.
The group wants governments to make defensive AI tools and authorised security testing available to organisations operating hospitals, water utilities and other critical services. It also calls on technology and cybersecurity companies to share expertise, threat intelligence and resources to help defenders identify and fix vulnerabilities.
For frontier AI developers, the letter specifically calls for responsible access to AI models, alongside funding, training and hands-on support for organisations that lack the resources to defend themselves against increasingly capable attacks.
The warning comes amid a series of incidents that have highlighted how AI systems can behave in unexpected ways when given access to tools and digital environments.
In July, a group of OpenAI agents being tested in a controlled environment reportedly created secret communication channels and worked together before successfully attacking Hugging Face. The incident has been described as an early example of an AI-enabled cyber-attack.
Other incidents involving AI systems developed by Anthropic and Meta have also raised concerns about AI agents finding ways around security safeguards, organising their activities and impersonating people.
The concern is not limited to AI-generated attacks. Earlier this week, the US Department of Justice said hackers linked to China had breached technology used by the US Senate, NASA, the Federal Reserve and the department itself.
The water sector has also faced mounting pressure. At least seven US water and wastewater companies have reported cyber-attacks, prompting the Federal Bureau of Investigation to issue an alert urging utilities to strengthen the security of their operational systems.
The same technology creating new risks is also being positioned as part of the answer. AI systems can scan software for weaknesses at a speed that would be difficult for human security teams to match, potentially allowing long-standing vulnerabilities to be found and fixed before attackers exploit them.
Anthropic has developed Mythos, a powerful system capable of identifying vulnerabilities that had reportedly escaped human detection for years. The company has restricted access to the technology because of concerns that its capabilities could also be misused.
OpenAI has launched its Daybreak programme, while Microsoft has introduced its Perception cyber platform, reflecting a wider push to apply advanced AI to defence.
The open letter argues that this defensive capability needs to reach organisations that are least equipped to build it themselves. It also calls for AI systems to become more observable, with AI agents identifiable and accountable for their actions.
The message is straightforward: AI may give attackers a powerful new edge, but it could give defenders one too. The signatories are urging governments and industry to build that defensive advantage now, rather than waiting for the next generation of AI-driven attacks to make the decision for them.