Five Ways To Prepare For AI-Accelerated Cyberattacks

🔍 Read the full analysis: Five Ways To Prepare For AI-Accelerated Cyberattacks on ThorstenMeyerAI.com

Age 18–24?Offer from Amazon

Prime made for students and young adults

• Fast, free delivery for dorm and study essentials
• Prime Video and Amazon Music included
• Member-only deals

Try Prime for Young Adults →Free trial for eligible 18–24 year olds

As an affiliate, we earn on qualifying purchases.Hide

U.S. agencies and Microsoft have reported AI-assisted cyber activity targeting industrial control systems and cloud organizations, including the use of generated scripts and an agentic framework. The reports do not show that AI makes attacks universally successful: Microsoft said most of the targeting in one campaign failed, but documented attempts at scale and, in several cases, access to privileged cloud accounts in under three hours.

A joint advisory published August 18 by the NSA, CISA, FBI, Department of Energy and Environmental Protection Agency says actors used AI-generated scripts against internet-reachable Siemens S7 programmable logic controllers, or PLCs. These devices control industrial processes. The agencies said the activity involved reconnaissance and capability development and assessed that it was likely preparation for operational effects against U.S. critical infrastructure. The advisory describes scripts disguised as legitimate monitoring tools and built using the open-source snap7 library.

In an October 21 report, Microsoft Threat Intelligence said the China-state-aligned actor Storm-0940 used ARTEX, an open-source multi-agent penetration-testing framework, in activity connected to the compromise of ConnectWise Cloud Backup. Microsoft said the actor tried to sign in to more than 2,250 tenants after obtaining credentials for the cloud backup app. Between October 10 and 14, Microsoft recorded more than 151,000 failed logons, peaking at 2,900 per minute; none of those attempts succeeded. In several attacks, Microsoft said, targeting to privileged cloud access took less than three hours.

Microsoft cautioned that most targeting was unsuccessful and that an affected app or successful sign-in does not by itself show that data was accessed. ConnectWise, which disclosed the compromise on October 16, said existing backup archives showed no indication of compromise and said it suspended Cloud Backup, SaaS Security and Cloud Management during its investigation. Separately, OpenAI said AI had identified critical PHP vulnerabilities patched on October 21 before attackers did, and forecast that similar findings would increase sharply in coming months. That is a forecast, not a measured estimate of future attack volume.

Five Defenses Against Faster Attacks

The reports point to a practical change in the pace and scale of routine attack work, not evidence of an all-powerful autonomous attacker. Software that can run continuously may shorten the time available to spot suspicious access, contain an account or respond to a vulnerable system. At the same time, tools that identify software flaws may generate more fixes and increase the pressure on organizations to apply them promptly.

Five steps follow from the specific risks described in the advisories and reports:

• Build an accurate inventory. Record internet-facing systems, cloud applications, service accounts and operational technology, including PLCs. Organizations cannot prioritize devices they do not know they have.
• Set and meet patch deadlines. Track vendor fixes for exposed systems and critical software, assign owners, and verify that updates were applied. The security value of a fix depends on deployment, not just publication.
• Strengthen sign-in controls. Use phishing-resistant multifactor authentication where available, especially for administrators and remote access. Review privileged accounts and remove access that is no longer needed.
• Review trusted third-party apps. Check what permissions backup and cloud-management applications hold, limit them to necessary access, and monitor sign-ins and unusual activity. A trusted connection can widen the impact of compromised credentials.
• Separate and practise response. Keep operational technology isolated from general IT where feasible, monitor control-system traffic, and rehearse how to contain compromised accounts or services. Incident plans should identify who can revoke app access and isolate affected networks.

These measures do not guarantee prevention. They address the exposures that the cited cases involved: reachable controllers, credentials, trusted application access and the time needed to detect and contain activity.

What the Recent Reports Document

The August advisory covers Siemens S7 PLCs, including models from the S7-200 through S7-1500, and identifies manufacturing, energy, water, chemical, food and agriculture, and commercial facilities as sectors of concern. The agencies recommend asset inventories, firmware updates, keeping PLCs off the public internet, perimeter restrictions on TCP port 102, separation between IT and operational technology, multifactor authentication for remote OT access and monitoring for unexpected write activity.

Microsoft’s account is tied to the ConnectWise Cloud Backup compromise, first disclosed by ConnectWise on October 16. Microsoft described ARTEX as a multi-agent framework capable of carrying out stages from reconnaissance through data exfiltration with limited human interaction. The reporting establishes that the framework was used in activity attributed by Microsoft to Storm-0940; it does not establish that every attempted tenant access succeeded or that customer data was taken.

The source material also describes a July 30 Unit 42 report about an actor using a model with the Hermes Agent framework. The available account is incomplete, so it does not support a detailed description of the agent’s actions or outcomes here. More broadly, one security leader’s forecast of a coming “vulnerability tsunami” and an approximately 18-month period of heightened pressure is an expert opinion, not a measured timeline.

“The majority of the targeting wasn’t successful.”

— Microsoft Threat Intelligence

Limits of the Current Evidence

The reports do not establish how often AI-assisted activity succeeds across the wider threat landscape, or how much of the observed activity would have been impossible without AI. In Microsoft’s account, unsuccessful logons are not evidence that tenant data was reached, and the company said a successful sign-in does not necessarily mean data was accessed. ConnectWise said its existing backup archives showed no indication of compromise; the source material does not provide a final investigation finding.

The agencies describe activity against PLCs and assess that it may be preparation for operational effects, but the material does not confirm that the reported activity caused physical disruption. OpenAI’s prediction of more AI-discovered vulnerabilities is forward-looking. The “vulnerability tsunami” and 18-month timeline cited by a security leader are opinion, not independently measured forecasts.

Monitoring Fixes and Investigations

Organizations should review the agencies’ PLC guidance and Microsoft’s account of the ConnectWise-related activity against their own systems, permissions and incident plans. ConnectWise’s investigation and any further findings about data access remain relevant to affected customers. Security teams should also track software updates and advisories as vulnerabilities are disclosed, then verify remediation rather than relying on announced fixes alone.

For individual users, the most direct starting point is securing the primary email account: enable multifactor authentication, use a unique password, review recovery options and check active sessions. The available reports do not provide a single forecast for how quickly AI-assisted attacks will spread. The next clear indicators will be additional technical findings, confirmed impact reports and evidence about whether organizations are closing the exposures identified in these cases.

Key Questions

Have AI-assisted cyberattacks been confirmed?

Government agencies and Microsoft have reported specific AI-assisted activity, including generated scripts aimed at industrial controllers and the use of an agentic framework in a cloud-related campaign. Those reports do not establish that all activity succeeded.

Did the ConnectWise-related campaign expose customer data?

The source material does not confirm that customer data was accessed. Microsoft said a successful sign-in does not necessarily indicate data access, and ConnectWise said existing backup archives showed no indication of compromise while it investigated.

What should companies do first?

Start by identifying internet-facing systems, cloud apps and operational technology. Then prioritize exposed-system patches, strong authentication, third-party app permissions and tested response procedures.

Are AI tools making every cyberattack more effective?

The reports support a narrower conclusion: AI tools were used in documented activity and can automate or accelerate some tasks. Microsoft said most targeting in the reported campaign was unsuccessful, so the material does not show that every AI-assisted attempt is effective.

Is a surge in AI-found vulnerabilities certain?

No. OpenAI forecast that findings would increase sharply in coming months, but that is a forward-looking statement. The cited source material does not provide a measured estimate of how many vulnerabilities will be found or how organizations will manage the resulting fixes.

Source: ThorstenMeyerAI.com