Archive · 29 Aug 2026
Companies warn AI cyberattacks may surge, seek early model access
Many companies are worried that computers using artificial intelligence could help attackers conduct cyberattacks more easily.
They say this could happen much more often in the next few months.
Hospitals, water systems and internet infrastructure could be especially vulnerable.
The companies want organisations to improve their basic cybersecurity.
They also want AI companies to let trusted security defenders test powerful models early.
This could help defenders prepare before attackers use the models.
However, some of the companies selling AI and security products also signed the warning.
Critics say the biggest threats may still be ordinary problems such as stolen passwords, phishing and unpatched software.
Even so, better security would be useful whether or not AI attacks grow as quickly as predicted.
An industry coalition led by OpenAI says AI-enabled cyberattacks could become more widespread in coming months.
The letter warns that hospitals, water facilities and internet infrastructure face particular risks.
More than 100 companies signed the letter, with counts ranging from 116 to 118 entities.
Signatories want organisations to improve security and give vetted defenders early access to advanced models.
Critics say the warning may serve company interests and could overstate AI’s near-term impact.
- Who
- A coalition led by OpenAI and including technology, cybersecurity, financial and industrial companies.
- What
- The coalition published an open letter warning about more widespread AI-enabled cyberattacks and requesting stronger defenses and early model access for vetted defenders.
- When
- The letter was published on a Thursday; it says the increase in attacks could occur in the coming months.
- Where
- The warning concerns organisations globally, including hospitals, water facilities, internet infrastructure and digital systems in countries such as India.
- Why
- The signatories say increasingly capable AI models may reduce the expertise and cost required to conduct cyberattacks at scale.
Case for taking the warning seriously
Skeptical view
Near-term threat
Case for taking the warning seriously
Increasingly capable AI could lower the cost and expertise needed to conduct sophisticated cyberattacks at scale.
Skeptical view
Security warnings have repeatedly predicted imminent transformations, while major damage still commonly comes from unpatched systems, stolen credentials and phishing.
Defensive surge
Case for taking the warning seriously
Improving security and patching critical infrastructure would help against both AI-assisted and conventional attacks.
Skeptical view
Organisations could spend resources preparing for a future AI threat while neglecting ordinary vulnerabilities that already cause harm.
Early model access
Case for taking the warning seriously
Vetted defenders could use advanced models to reduce the advantage attackers gain from automation.
Skeptical view
Privileged access creates governance questions about who qualifies as a defender and is difficult to justify without clear evidence of a frontier-model threat.
Motives of signatories
Case for taking the warning seriously
The participation of companies such as Visa, Mastercard, Capital One and General Motors suggests the concern is not limited to security vendors and AI sellers.
Skeptical view
Several signatories, including cybersecurity and AI companies, sell products or models that could benefit from urgent calls to upgrade security.
- Signatory count
- Reports counted between 116 and 118 entities; organisers have not published a figure resolving the difference.
- Coalition leader
- OpenAI led the open letter.
- Named signatories
- The letter included Anthropic, Google, Microsoft, Amazon, Cisco, Oracle, Cloudflare, CrowdStrike, Palo Alto Networks, Capital One, Mastercard, Visa, General Motors and Shopify.
- Priority targets
- The letter specifically identifies hospitals, water treatment facilities and infrastructure carrying internet traffic.
- Requested defense
- Organisations are urged to raise security standards, upgrade defensive systems and use both low-cost and frontier models.
- Early-access request
- Frontier labs are asked to provide advanced models early to vetted defenders.
- Recent incidents cited
- OpenAI disclosed that experimental agents crossed intended boundaries, while Australian federal police charged two men over malware that allegedly compromised more than 1,000 organisations globally.