How Companies And Users Can Respond To AI-Powered Cyberattacks

🔍 Read the full analysis: How Companies And Users Can Respond To AI-Powered Cyberattacks on ThorstenMeyerAI.com

Buying for a business?Offer from Amazon

Get business pricing on tech for your team

• Business-only prices and quantity discounts
• Tax-exempt purchasing
• Multiple users, one account, clear invoices

Create a free Business account →Free to register

As an affiliate, we earn on qualifying purchases.Hide

US agencies and cybersecurity researchers have documented attackers using AI-generated scripts and AI agents in activity targeting industrial controllers and cloud tenants, giving companies and individual users new reasons to review established defenses. The reports describe faster automation of familiar attack steps—not proof of an autonomous system capable of succeeding against every target—and note that many attempts were unsuccessful.

A joint advisory issued August 18 by the NSA, CISA, FBI, Department of Energy and Environmental Protection Agency says threat actors used AI-generated exploitation scripts against internet-reachable Siemens S7 programmable logic controllers, or PLCs. These industrial computers control equipment and processes. The agencies say the activity included reconnaissance and capability development against US critical infrastructure, and assess that it may be preparation for operational effects. They identify manufacturing, energy, water, chemical, food and agriculture, and commercial facilities among the sectors targeted.

On October 21, Microsoft Threat Intelligence reported that the actor it calls Storm-0940 used ARTEX, an open-source multi-agent penetration-testing framework, in activity connected by Microsoft to the ConnectWise Cloud Backup compromise. Microsoft says the actor used credentials associated with the backup app to attempt sign-ins to more than 2,250 tenants. A password-spraying campaign from October 10 to 14 generated more than 151,000 failed logons, peaking at 2,900 per minute; Microsoft said none of those attempts succeeded. In several attacks, Microsoft said, the interval from initial targeting to privileged cloud access was under 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 said its existing backup archives showed no indication of compromise and that it suspended Cloud Backup, SaaS Security and Cloud Management while investigating. Separately, OpenAI said AI had identified critical PHP vulnerabilities before they were patched on October 21 and forecast that AI-discovered critical vulnerabilities will “increase sharply in the coming months.” That is a forecast, not a count of future flaws.

Faster Attacks Raise the Cost of Delay

The reports matter because AI can reduce the time and labor needed for repeated tasks such as reconnaissance, scripting and trying credentials across many accounts. That can compress the window defenders have to notice suspicious activity and respond. It does not mean attackers have stopped making mistakes or that automated attempts routinely work: Microsoft explicitly said most targeting failed, and its logon figures describe failed attempts.

For companies, the practical consequence is operational. A compromised supplier account or trusted application can offer a route into customer environments, while exposed industrial controllers can put physical processes at risk. Organizations need to know which systems are reachable, which third parties hold access and who can isolate affected services. For individuals, email and other identity accounts remain important because attackers can use compromised credentials to reset access to additional services.

AI may also increase the volume of vulnerabilities defenders need to assess. OpenAI’s prediction of a sharp rise is not evidence that every flaw will be exploitable, but it points to greater pressure on teams to prioritize verified fixes. Inventory, timely patching, strong authentication and practiced response plans remain useful whether an attacker uses AI or conventional tools.

Two Documented Routes Into Systems

The August advisory describes activity against Siemens S7 PLCs, including models from the S7-200 through the S7-1500. The agencies say actors found exposed devices using internet-scanning services and used scripts disguised as legitimate monitoring tools, built with the open-source snap7 library. They characterize this as lowering the expertise and time required for the activity. Their recommendations include keeping controllers off the public internet, inventorying and patching them, separating operational technology from business networks, requiring multifactor authentication for remote access and monitoring for unauthorized S7 write activity.

Microsoft’s October report describes a different route: a trusted cloud backup application and its credentials. ConnectWise disclosed the compromise on October 16; Microsoft later detailed Storm-0940’s use of ARTEX. The incident highlights how third-party software access can extend risk beyond the supplier itself. Microsoft has described Storm-0940 as a China-state-aligned actor active since 2021; that attribution is Microsoft’s assessment.

A separate July 30 report from Palo Alto Networks’ Unit 42 described an actor using the DeepSeek model with the Hermes Agent framework, controlled through Telegram. Unit 42 said the agent performed extensive targeting analysis quickly. These reports describe distinct investigations and should not be treated as proof that all AI-assisted activity is connected or uses the same tools.

“This is not a theoretical risk—it is an active threat.”

— The joint NSA, CISA, FBI, DOE and EPA advisory

What the Reports Do Not Establish

The reports do not establish how often AI-assisted attacks succeed across all organizations, or whether AI was necessary for the reported outcomes. Microsoft said the majority of targeting failed and that successful sign-in does not necessarily mean data was accessed. ConnectWise’s statement about its backup archives is not a finding that every affected service or customer environment was unaffected; the company said it had suspended services while investigating.

The agencies assess that PLC activity is likely preparation for operational effects, but the supplied advisory details do not confirm that the activity caused physical disruption. Nor is OpenAI’s forecast a verified estimate of how many vulnerabilities will be found or exploited. The scope, duration and results of ongoing investigations remain unclear.

Actions Before the Next Incident

Companies should begin with an inventory of internet-facing assets, industrial controllers, cloud applications and third-party access. For operational technology, follow the agencies’ guidance: remove public exposure, patch firmware where safe, segment OT networks from IT, use multifactor authentication for remote access and monitor for unexpected controller writes. Any network changes should be planned with operational safety and service continuity in mind.

For cloud and business systems, review which applications have access to tenant data, revoke credentials that are no longer needed and investigate unusual sign-ins or privilege changes. Set patch deadlines based on risk and exposure, and test incident-response procedures—including who can disable an integration or isolate a system. Boards should ask whether these controls are in place rather than treating AI adoption itself as the main measure of readiness.

Individual users can protect their primary email account with a unique password and phishing-resistant multifactor authentication where available. They should review account recovery settings and connected apps, and report unexpected sign-in alerts rather than approving prompts they did not initiate. Security teams should watch for further agency advisories, vendor investigation updates and revised guidance as the reported incidents develop.

Key Questions

Does this mean AI can independently hack any company?

No. The reports describe AI-generated scripts and agents helping with specific tasks. They do not show that AI can reliably compromise any target. Microsoft said most of the targeting in its report was unsuccessful.

What should a company address first?

Identify exposed systems and third-party access, patch known vulnerabilities, require multifactor authentication for remote access and review unusual account activity. Organizations with industrial systems should make sure controllers are not reachable from the public internet.

Was customer data accessed in the ConnectWise incident?

The supplied reports do not establish that all affected customers’ data was accessed. Microsoft said an affected app or successful sign-in does not necessarily mean data was accessed. ConnectWise said existing backup archives showed no indication of compromise while it investigated.

What can individual users do?

Secure the email account used to recover other accounts with a unique password and phishing-resistant multifactor authentication where available. Review recovery details and connected apps, and do not approve sign-in requests you did not initiate.

Source: ThorstenMeyerAI.com