On September 22, someone inside a US retailer’s network typed “pwoershell.” Then “systemifo.” Then they ran the same blocked loader five times in about 20 minutes, getting stopped by endpoint security each time.1

It wasn’t a sophisticated operation. It was someone stuck.

The next morning, the typos were gone. Over roughly two hours, about 50 long, clean PowerShell scripts ran on the same server, complete with labeled output, error handling, and comments that read like a model thinking out loud. About ten minutes after the attackers first queried the retailer’s SolarWinds Orion service, they used its bundled messaging software to run commands on a second server. Symantec’s researchers, who traced the intrusion, put it plainly: “The day before, the same operators with the same access had achieved little.”1,2

That line stuck with us, because it captures the last two weeks of security news better than any headline did.

Same access, different results

The retailer story is worth reading closely, because the details are so ordinary. The attackers got in through SQL Server hosts, harvested saved credentials from admin tools, and tested whether the servers could reach the internet. Later, they scanned more than a dozen internal hosts for SQL, SMB, RDP, and WinRM, and ran a password sprayer from yet another machine. Symantec saw no data theft or encryption, but the attackers did collect domain credentials.1

“None of the techniques used in this second phase of the attack were new,” Symantec wrote. “What changed was how quickly the attackers identified a complex workaround and how reliably it was applied.”1

The paths were already there. The AI just found them faster than a tired human could.

A pattern across the headlines

That theme kept surfacing in Dark Reading’s coverage over the past two weeks.

On Nov. 3, the UK’s National Cyber Security Centre published its State of the Cyber Nation report. It warned that AI-powered malware is “highly likely” to be used in a ransomware attack on UK critical infrastructure by 2028. The more striking point was the NCSC’s explanation for the rise in incidents: long-tolerated vulnerabilities being found and exploited at scale, not AI inventing new methods. NCSC chief executive Richard Horne didn’t soften it: “For decades, we have collectively got away with doing tech on the cheap.”3,4

Three days later, Japan announced a 10 billion yen emergency cybersecurity package after a wave of breaches, citing “AI-generated attacks” as the reason to move fast.5,6

And Check Point Research detailed how criminals turned ARTEX, an open-source AI penetration-testing framework, into an attack platform. Its agents plan, scan, probe, and document on their own. In one campaign against 292 cryptocurrency-related domains, they found an unauthenticated API that exposed AWS credentials with write access to a storage bucket.7,8 Separately, hundreds of servers running the tool were found sitting open on the internet with no authentication at all.9

Read together, these stories aren’t really about AI. They’re about exposure that organizations have lived with for years, and how little time is left to find it first.

What defenders can control

You can’t control how fast an attacker’s model writes PowerShell. You can control what that code is able to reach once it runs.

Walk back through the retailer intrusion with that lens. Could the SQL Server hosts reach the internet, and did they need to? Could a web server talk directly to the monitoring platform’s message broker? Could one compromised machine scan a dozen others over SQL, SMB, RDP, and WinRM? Every one of those is a question about network paths, and every answer lives in a firewall rule, a cloud security group, or a segmentation policy somewhere.

Most security teams can’t answer those questions quickly. The answer is usually spread across devices from several vendors, years of rule changes, and exceptions nobody remembers approving.

“When I read about that retailer, I don’t think about the AI. I think about the paths,” says Jeffrey Spear, CISO at Tufin. “The SQL servers that could reach the internet, the one machine that could scan a dozen others: every one of those moves was something the network allowed. As a CISO, that’s the part I can actually close, but only if I can see it first.”

This is the problem Tufin works on with its customers every day. Tufin Orchestration Suite builds a single, live view of connectivity across firewalls, clouds, SASE, and microsegmentation, so a team can ask “what can reach this server?” and get an answer in minutes instead of days. Segmentation Intelligence checks continuously for the drift that turns a clean design into a flat network: zones that have grown overcrowded, subnets that never got assigned, rules that quietly cross a boundary they shouldn’t. And when something needs to change, Tufin Network Security Agents can design and check the change across the whole network, while your team keeps the final say on anything risky.

None of that stops an attacker from getting a foothold. It shrinks what the foothold is worth.

The quiet takeaway

It’s tempting to read two weeks of AI attack headlines and conclude that defense needs to be reinvented. The evidence points somewhere more practical. The retailer’s attackers used old techniques. The NCSC blamed old vulnerabilities. The ARTEX campaigns found exposed APIs and weak admin pages.

What AI changed is the clock. The work of cleaning up access paths, tightening segmentation, and knowing what’s reachable used to be something teams could get to next quarter. Now it’s the work that decides whether a stuck attacker stays stuck.

Works Cited

  1. Symantec Threat Hunter Team. “How AI Revived a Stalled Attack on a U.S. Retailer.” Security.com, Broadcom, 12 Nov. 2026, www.security.com/blog-post/ai-assisted-attack-us-retailer.
    1. Culafi, Alexander. “Inside an AI Attack Against a US Retailer.” Dark Reading, 12 Nov. 2026, www.darkreading.com/cyberattacks-data-breaches/inside-ai-attack-us-retailer.
    2. National Cyber Security Centre. “Cyber Chief: UK Facing AI-Powered Attacks on Critical Infrastructure in Two Years.” NCSC, 3 Nov. 2026, www.ncsc.gov.uk/news/cyber-chief-uk-facing-ai-powered-attacks-on-critical-infrastructure-in-two-years.
    3. “UK Warns Its Critical Infrastructure Faces Grave Risk.” Dark Reading, Nov. 2026, www.darkreading.com/cybersecurity-operations/uk-warns-critical-infrastructure-grave-risk.
    4. “Japan Announces Emergency Cybersecurity Package Following a Wave of Attacks.” Associated Press, 6 Nov. 2026. KPRC Click2Houston, www.click2houston.com/tech/2026/11/06/japan-announces-emergency-cybersecurity-package-following-a-wave-of-attacks/.
    5. “Japan Pushes Emergency Funding for Cybersecurity as AI-Assisted Breaches Rise.” Dark Reading, Nov. 2026, www.darkreading.com/cybersecurity-operations/japan-emergency-funding-cybersecurity-ai-assisted-breaches.
    6. Ofonagoro, Joseph Chisom. “ARTEX Cyberattacks: AI Agents Steal Credentials and Signing Certificates.” eSecurity Planet, 9 Nov. 2026, www.esecurityplanet.com/threats/news-artex-ai-agents-credential-theft/.
    7. “Artex: The AI Pen-Test Tool Hackers Turned into a Weapon.” Dark Reading, Nov. 2026, www.darkreading.com/cyberattacks-data-breaches/artex-ai-pen-test-tool-hackers-weapon.
    8. “Hundreds of Artex AI Hacking Tools Left Wide Open Online.” TechBuzz, 6 Nov. 2026, techbuzz.ai/articles/hundreds-of-artex-ai-hacking-tools-left-wide-open-online.

    Ready to Learn More

    Get a Demo