What Companies Need To Know About AI-Accelerated Cyberattacks

🔍 Read the full analysis: What Companies Need To Know About AI-Accelerated 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 Microsoft have reported separate cases of AI-assisted cyber activity involving industrial control systems and cloud tenants, including the use of AI-generated scripts and an agentic framework. The reports show how automation can speed routine attack steps, but they also describe failed attempts and do not establish that all targeted systems were compromised.

A joint advisory published 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. PLCs are industrial computers used to control equipment and processes. The agencies say the activity involved reconnaissance and capability development and assess that it may be preparation for operational effects against critical infrastructure.

Microsoft Threat Intelligence reported on October 21 that the China-state-aligned actor it calls Storm-0940 used ARTEX, an open-source multi-agent penetration-testing framework, in activity linked by Microsoft to the ConnectWise Cloud Backup compromise. Microsoft says the actor used credentials associated with the app to attempt sign-ins to more than 2,250 tenants. From October 10 to 14, Microsoft recorded more than 151,000 failed logons, with a peak of 2,900 per minute; none of those attempts succeeded. In some attacks, Microsoft says, the time from initial targeting to 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 mean data was accessed. ConnectWise disclosed the compromise on October 16 and said existing backup archives showed no indication of compromise. It suspended Cloud Backup, SaaS Security and Cloud Management while investigating. Separately, OpenAI said AI had identified critical PHP vulnerabilities patched on October 21 before attackers did, and forecast that AI-found critical vulnerabilities would increase sharply in the coming months. That is a prediction, not a measured trend.

Faster Attacks Raise the Cost of Delay

The reports matter because automated reconnaissance, credential attempts and exploit adaptation can compress work that would otherwise take longer. Microsoft says agents can operate continuously, while Palo Alto Networks’ Unit 42 reported that an agent completed hundreds of hours of manual targeting analysis in minutes. These are source descriptions of particular capabilities, not proof that every criminal group has the same tools or results.

For companies, the practical risk is that defenders may have less time between an exposed weakness and an attempted intrusion. This is especially relevant where industrial equipment is reachable from the internet or where a trusted third-party application has access across many customer tenants. At the same time, the Microsoft figures show why raw activity should not be confused with successful compromise: the reported password-spray attempts failed, and Microsoft says most targeting was unsuccessful.

AI can also help defenders identify flaws. OpenAI’s account of vulnerabilities found before attackers did illustrates that the same broad technology can support security research and remediation. If the volume of discovered flaws rises, however, companies may face more pressure to triage and patch quickly. The scale and timing of any such increase remain uncertain.

Two Reports, Different Attack Paths

The August advisory focuses on operational technology: Siemens S7-200 through S7-1500 controllers, including safety controllers, in sectors such as manufacturing, energy, water, chemicals, food and agriculture, and commercial facilities. The agencies say actors found exposed PLCs through internet scanning services and used scripts built with the open-source snap7 library, disguised as monitoring tools. They describe this as reducing the technical expertise and time needed for the activity.

The Microsoft report concerns a different route: trust in a cloud application. Microsoft says Storm-0940 used credentials associated with ConnectWise Cloud Backup and ARTEX to attempt access across customer tenants. ConnectWise’s service suspension and investigation followed its October 16 disclosure. These accounts should not be merged into one campaign: they describe distinct activity, targets and evidence.

A third example in the source material comes from Unit 42’s July 30 report. It describes a Chinese-speaking actor using the DeepSeek model with the Hermes Agent framework and directing it through Telegram. The supplied account says the agent searched GitHub after its initial exploit failed. This illustrates experimentation with agent-based workflows, but the available material does not establish the operation’s broader impact or connect it to the other reported cases.

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

— Joint Cybersecurity Advisory by the NSA, CISA, FBI, DOE and EPA

Impact and Future Scale Remain Unknown

The reports do not establish how much data, if any, was accessed through the ConnectWise-related activity. Microsoft says a successful sign-in does not necessarily indicate data access, and ConnectWise said its existing backup archives showed no indication of compromise. The companies’ investigations and the full scope of affected customers were not resolved in the supplied material.

The agencies describe PLC activity and assess it as likely preparation for operational effects, but the source material does not confirm that the reported activity caused physical disruption. It also does not give a count of affected industrial installations. OpenAI’s prediction of a sharp rise in AI-found vulnerabilities and an expert’s forecast of a near-term “vulnerability tsunami” are forward-looking opinions, not established outcomes or quantified forecasts.

More broadly, the reports do not show that AI alone caused these incidents or that automated attacks are uniformly more successful. The exact degree of human involvement, the prevalence of these methods across attackers, and how quickly defensive tools will adapt remain unclear.

Prioritize Controls and Track Findings

Companies should follow the advisory and vendor updates as investigations proceed, then check whether their own systems share the reported exposure patterns. For operational technology, the agencies recommend inventorying controllers, patching firmware and preventing direct internet access. They also advise separating IT and OT networks with a demilitarized zone, using one-way gateways for historian connections where appropriate, requiring multifactor authentication for remote OT access, and monitoring S7 traffic for writes outside approved change windows.

For cloud services, security teams should review third-party application permissions, credentials and tenant sign-in logs, and confirm how vendors notify customers about incidents. Microsoft and ConnectWise had not provided, in the supplied material, a final accounting of the investigation’s scope or data impact. Organizations that use the affected services should rely on vendor guidance and their own evidence rather than assume either compromise or safety from a sign-in alert alone.

Across both environments, the source material points to established controls rather than a need to buy AI products: maintain an asset inventory, set patch deadlines, use phishing-resistant authentication where available, review app access, segment networks and rehearse incident response. Individuals can start by securing their email account, which often serves as a recovery and access channel for other services. The next meaningful developments are the results of ongoing investigations and any further government, vendor or vulnerability disclosures.

Key Questions

Have AI-assisted attacks been confirmed?

Government and industry reports describe specific AI-assisted activity, including AI-generated scripts against Siemens PLCs and Microsoft-reported use of the ARTEX framework. Those reports do not show that every detail of every incident is independently confirmed.

Were all 2,250 targeted cloud tenants compromised?

No. Microsoft reported attempts to sign in to more than 2,250 tenants, but said most targeting was unsuccessful. It also cautioned that a successful sign-in does not necessarily mean data was accessed.

Did the PLC activity disrupt industrial operations?

The supplied advisory describes reconnaissance and capability development and says the activity may be preparation for operational effects. It does not confirm that the activity caused physical disruption.

What should companies do first?

Start with asset and access basics: identify exposed systems, patch known weaknesses, restrict internet access to industrial controllers, review third-party app permissions, strengthen remote sign-in and practise incident response.

Source: ThorstenMeyerAI.com