For years, cybersecurity professionals have warned about the growing threat of AI use in cyberattacks—be it using AI writing tools to generate more convincing phishing emails or using voice mirroring and AI-generated video technology to trick victims into clicking malicious links or sending money.

But as AI is swiftly incorporated into our everyday lives, recent security evaluations from OpenAI and Anthropic suggest there’s a larger, more complex threat starting to emerge in the cybersecurity space.

See also: AI is superpowering cyberattacks, but manufacturers can cut their exposure

AI is beginning to coordinate entire attack sequences, linking together reconnaissance, privilege escalation, lateral movement and persistence without a human directing every individual action.

Today, AI can compress attacks that once took months into days, fundamentally changing how organizations must think about cyber defense, and how quickly they need to adapt to address the growing capabilities of bad actors.

The relevance of this trend to manufacturers, who are starting to widely pilot or operationalize agentic AI, cannot be understated. Nearly three in four companies are planning to deploy agentic AI within two years—and one in five reported having an equipped model, according to Deloitte’s 2026 State of AI in the Enterprise.

Where agentic AI is emerging as an ally, it’s also being weaponized as a threat to factory IT and OT operations.

A wake-up call for manufacturing IT and OT

In July, OpenAI disclosed that one of its advanced AI models escaped its testing environment, chained together multiple attack techniques, and ultimately compromised Hugging Face's production infrastructure while attempting to complete a cybersecurity benchmark.

That's what enables attacks today—which may have once unfolded over weeks or months—to progress in a matter of days, and it’s why organizations need to rethink how they approach cybersecurity defense.

Equally important is persistence. Unlike a human attacker, an AI-powered hacking system doesn't stop at the end of the workday or wait for another specialist to take over. AI can continue testing different attack paths, adapt when one approach fails and keep working toward its objective for as long as its computing resources allow.

That persistence, combined with the ability to run multiple processes simultaneously, is a key reason these attacks can unfold so much faster than traditional operations.

What does this mean for manufacturers?

Manufacturing organizations can be attractive targets for cyberattacks because they sit at the intersection of valuable intellectual property, connected production environments and increasingly digital operations.

Engineering data, OT, supplier networks and cloud-based business systems all create opportunities for attackers once they've established an initial foothold. The biggest recent shift in cybercrime isn't necessarily more sophisticated attacks, but faster, more coordinated ones that reduce defenders' response time.

See also: AI stokes debate over cloud-powered compute vs. on-prem

Historically, security teams could often identify suspicious activity as attackers moved through different stages of an operation, but agentic AI compresses that timeline.

When reconnaissance, credential compromise, privilege escalation and lateral movement become part of a single autonomous workflow, defenders have far less time to detect and interrupt an attack before it affects business operations.

Manufacturers should assume sophisticated attackers are increasingly incorporating AI into their operations.

That doesn't mean every attack will be fully autonomous, but as agentic systems continue to evolve, AI is becoming another force multiplier capable of accelerating attacks while adapting to changing conditions—and attackers will look for ways to use them.

Defending against AI-powered cyberattacks

The rise of agentic AI doesn't change the fundamentals of cybersecurity. If anything, it reinforces why they matter and the importance of, to use an American football analogy, getting the blocking and tackling right.

Least-privilege access, network segmentation, credential management and layered security controls remain some of the most effective ways to interrupt an attack before it progresses through the cyber kill chain.

Every additional checkpoint forces an attacker to solve another problem and creates another opportunity to detect suspicious behavior before meaningful damage occurs. While agentic AI can compress the timeline of an attack, it doesn't eliminate those opportunities.

See also: Black Kite: Ransomware increasing across all metrics in 2026

Every phase of the cyber kill chain remains a potential point of interruption. The challenge for manufacturers is to ensure there are enough defensive layers in place to stop an attack before it reaches its objective.

Organizations should also begin treating AI as part of their defensive strategy. AI-powered security tools can continuously monitor environments, identify anomalies and interrupt attacks before they progress through the kill chain. As attackers increasingly use AI to operate at machine speed, defenders will need similar capabilities to keep pace.

The goal isn't to prevent every attempted intrusion, but to create enough defensive checkpoints that even a persistent, autonomous attacker can't move seamlessly from one phase of an attack to the next.

The next AI arms race

The recent OpenAI and Anthropic evaluations demonstrate that AI-enabled cyber threats are no longer a future concern, but real problems we need to address today.

The good news is that the same technology reshaping cyberattacks can also strengthen cyber defense. AI can help organizations identify vulnerabilities earlier, detect unusual behavior faster and respond before an attacker completes the cyber kill chain.

Manufacturers that proactively strengthen security practices and begin incorporating AI into their defensive strategies will be better positioned as this new era of cybersecurity continues to evolve.