Read the full analysis: AI Is Accelerating Cyberattacks: Steps To Protect Your Business And Data on ThorstenMeyerAI.com
TL;DR
U.S. agencies say attackers used AI-generated scripts against internet-reachable industrial controllers, while Microsoft links an AI agent framework to activity involving a compromised cloud backup app. Microsoft says most targeting attempts failed, and a successful sign-in does not by itself prove data access. The reports point to practical priorities: secure exposed systems, review trusted applications, strengthen sign-in protections and rehearse incident response.
U.S. agencies and Microsoft have reported documented cases of attackers using AI-assisted tools against industrial control systems and cloud environments, adding evidence that software agents can speed up familiar cyberattack steps. The reports describe different incidents and do not establish that AI makes attacks reliably successful: Microsoft said most of the targeting it observed failed, and that a successful sign-in does not necessarily mean data was accessed.
In an August 18 joint advisory, the NSA, CISA, FBI, Department of Energy and Environmental Protection Agency said actors were conducting reconnaissance and developing capabilities against Siemens S7 programmable logic controllers, or PLCs, used in industrial settings. The agencies said attackers found internet-exposed devices and used AI-generated exploitation scripts disguised as monitoring tools, built with the open-source snap7 library. They assessed the activity as likely preparation to cause operational effects, not proof that such effects had occurred.
Microsoft Threat Intelligence reported on October 21 that it linked the China-state-aligned actor Storm-0940 to use of ARTEX, an open-source agentic penetration-testing framework, in activity involving the ConnectWise Cloud Backup compromise. Microsoft said the actor used credentials for the cloud backup application to attempt logins to more than 2,250 customer tenants. From October 10 to 14, Microsoft recorded more than 151,000 failed logons, with a peak of 2,900 per minute; it said none of those attempts succeeded. In several attacks, Microsoft said, the interval between initial targeting and privileged cloud access was under three hours.
Microsoft cautioned that most targeting was unsuccessful and that an affected application or successful sign-in does not necessarily show that data was accessed. ConnectWise, which disclosed the compromise on October 16, said existing backup archives showed no indication of compromise. It suspended Cloud Backup, SaaS Security and Cloud Management while investigating. These qualifications matter: the reports establish attempted activity and some access, but not confirmed data theft in every case.
Primary sources · Jul–Oct 2026
Infographic 01 · The record so far
AI is in the attacker’s toolchain. On the record.
Six documented events in twelve weeks. Each label says what the source itself states about AI. Figures are the vendors’ and agencies’ own.
Most AI-driven attempts in these reports failed. What changed is the cost of trying.Assessment: Thorsten Meyer
Sources: Unit 42 (30 Jul 2026); NSA, CISA, FBI, DOE, EPA advisory (18 Aug 2026); Microsoft Threat Intelligence (21 Oct 2026); ConnectWise Trust Center (from 16 Oct 2026); OpenAI, Patch the Planet (21 Oct 2026); Quest Diagnostics Form 8-K (22 Oct 2026).
Powered by Thorsten Meyer AI
Why Faster Attacks Test Basic Defenses
The reports suggest AI can reduce the time and expertise needed for parts of an attack, including reconnaissance, script development and repeated login attempts. Microsoft said ARTEX can operate with limited human interaction, while the agencies described AI-generated scripts aimed at industrial controllers. That does not mean attackers have gained a dependable way past defenses. In Microsoft’s account, a large volume of password-spray attempts failed, underscoring that AI-assisted activity is not the same as a successful breach.
For businesses, the practical concern is the speed and scale of routine risks. A stolen credential or overly trusted third-party application can expose many accounts before a response team has time to investigate. In industrial environments, an internet-accessible controller can create risks to operations as well as information. The reported activity makes timely patching, strong authentication and clear boundaries between corporate IT and operational technology more pressing.
Security leaders should treat the reports as a reason to review controls, not as evidence that every organization faces an imminent AI-driven attack. The source material includes a forecast from an unnamed security leader of a coming “vulnerability tsunami” and an estimated 18-month period to respond. That is an expert opinion, not a measured projection or established timeline. Organizations can act on the documented risks without relying on that forecast.
Two Reports, Different Attack Paths
The government advisory and Microsoft report describe separate activity. The first concerns Siemens S7 PLCs and possible preparation for operational effects at U.S. critical-infrastructure sites. The second concerns Storm-0940, a cloud backup application and attempts to access customer tenants. They should not be treated as a single campaign or as evidence that the same tools or operators were involved.
The government agencies recommended inventorying controllers, updating firmware and preventing PLCs from being reachable from the public internet. Their recommendations also include separating IT and operational technology networks, requiring multifactor authentication for remote OT access and monitoring industrial protocol traffic for unauthorized writes. For cloud services, Microsoft’s account highlights the importance of reviewing third-party application permissions and investigating unusual sign-in activity. These are established defensive measures; the reports do not show that buying a particular AI security product is necessary.
A separate OpenAI statement in the source material said AI had identified critical PHP vulnerabilities that were patched on October 21, before attackers did, and predicted that similar findings would rise sharply in coming months. That is a company statement and forecast about defensive vulnerability discovery, not evidence that attackers have already exploited those flaws. More discoveries could increase pressure on software maintainers and customers to assess and apply fixes promptly.
“The majority of the targeting wasn’t successful.”
— Microsoft Threat Intelligence
What Investigators Have Not Established
The public reporting does not establish whether the activity against Siemens controllers caused operational disruption or how many facilities were affected. The agencies described likely preparation for operational effects, a stated assessment rather than confirmation of an attack causing damage. The available material also does not identify individual organizations targeted in that activity.
For the ConnectWise-related activity, Microsoft said a successful sign-in does not necessarily mean data was accessed, and ConnectWise reported no indication of compromise in existing backup archives. The full scope, any data access and the final investigation findings remain unclear in the material provided. Microsoft’s attribution to Storm-0940 is its assessment, not an independently established finding in the source material.
The source also refers to a Unit 42 report about an actor using a language model with an agent framework, but the supplied account ends before describing the incident’s outcome. No further claims about that case can be confirmed here. More broadly, the reports do not quantify how often AI is used in attacks or show that it is the cause of a particular incident simply because attackers used automated tools.
Investigations and Security Reviews Continue
ConnectWise suspended Cloud Backup, SaaS Security and Cloud Management while it investigated. The source material does not give a completion date or describe the final findings, so affected customers should follow the company’s incident updates and review relevant account activity and application access.
Organizations using industrial controllers can compare their exposure and controls with the joint advisory’s recommendations, including checking whether PLCs are internet-reachable, applying firmware updates and limiting remote access. Cloud customers can review third-party app permissions, strengthen authentication and investigate unusual sign-in patterns. Boards and security teams can request current asset inventories, patch timelines and incident-response exercises rather than treating AI adoption itself as the only security issue.
Further agency advisories, company investigation updates and vulnerability disclosures may clarify the extent of these cases. Until then, the confirmed reports support focused defensive work, while broader claims about a coming surge in attacks remain forecasts.
Key Questions
Have AI-assisted attacks caused confirmed damage in these reports?
The joint advisory describes activity against Siemens controllers and assesses that it may be preparation for operational effects, but the supplied material does not confirm damage. Microsoft reported many failed login attempts and cautioned that a successful sign-in does not necessarily mean data was accessed.
What did Microsoft report about Storm-0940?
Microsoft said Storm-0940 used the ARTEX agentic framework in activity involving the ConnectWise Cloud Backup compromise. It reported attempted access to more than 2,250 tenants and more than 151,000 failed logons from October 10 to 14; Microsoft said none of those password-spray attempts succeeded.
What should businesses do first?
Start with an accurate inventory of internet-facing systems and third-party applications. Apply available patches, use multifactor authentication for remote access, review application permissions, separate industrial networks from corporate IT and practise incident response.
Does the vulnerability forecast mean a surge is certain?
No. OpenAI said it expects AI-found critical vulnerabilities to increase sharply, and an unnamed security leader forecast a period of heightened vulnerability pressure. Both are forecasts; the source material does not establish their future scale or timing.
Source: ThorstenMeyerAI.com