What Is AI Penetration Testing? The Hired Attack, Explained

A penetration test is a hired attack: security testers try to break into a company’s own systems to find the holes before a criminal does. Vendors such as Terra and OpenAI now sell AI agents that do much of that work, per their own product pages.

Palo Alto Networks is in talks to buy Terra Security, an AI pen-testing startup, in a deal that could top $500 million, Calcalist reported, as we covered in Palo Alto’s Terra deal.

Separately, more than 50 leaders told Axios they expect constant cyberattacks over the next 18 months, as we covered in the 18-month warning. Below: what a pen test is, what AI changes, and why buyers pay.

Business Pill · Least Privilege

An evergreen Business Engineer explainer. It teaches the general idea only and does not discuss this case.

The key insight: AI penetration testing hands hired attacks on a company’s own systems to software agents, and Terra says certified pentesters sign its reports; card-industry rules require internal and external tests at least once every 12 months, and Palo Alto may pay over $500 million, about 13 times what Terra raised, for one maker, Calcalist reported.

What Is a Penetration Test?

The US standards body NIST defines it in Special Publication 800-115: “security testing in which assessors mimic real-world attacks to identify methods for circumventing the security features of an application, system, or network.”

NIST adds that it “often involves launching real attacks on real systems and data”, and that most tests look for combinations of weaknesses that together give more access than any single one.

PCI DSS treats scans and tests as separate duties: internal vulnerability scans “At least once every three months”, and penetration tests on their own schedule, per the requirement text quoted below.

Terra Security raised $38 million ($8 million seed, $30 million Series A), per the company and SecurityWeek; C
Terra Security raised $38 million ($8 million seed, $30 million Series A), per the company and SecurityWeek; Calcalist reported Palo Alto Networks is in talks at a price that could exceed $500 million, about 13 times what Terra raised by our arithmetic. No deal has been announced.

What Changes When AI Runs It

Terra says its platform “unifies continuous agentic pentesting across AI systems, network infrastructure, and web apps”, on its homepage. Its agents, it says, “are trained on your business context and code”.

The company claims its approach is faster than traditional pentesting, “from 4-6 weeks to 2-4 hours”. That is Terra’s own claim; we have not seen an independent test of it.

OpenAI’s Daybreak Red Team page calls its product “AI penetration testing for web applications”, with agents that test apps, validate vulnerabilities and check whether a patch stops the exploit, as we covered in OpenAI’s pen-testing launch. It is in early access and lists no price.

Flow diagram: how an AI penetration test works, from scope and AI agent scanning to validation, human review, report and fix
How an AI penetration test works: the company sets the scope, AI agents map and test apps, networks and AI systems, weaknesses are chained and validated, people review at critical points and certified testers sign the report, per Terra; PCI DSS v4.0 requires internal and external tests at least once every 12 months, and NIST SP 800-115 warns testing can damage systems.

Where the Humans Stay

Terra keeps people in the process. It describes “Human-on-the-Loop at critical decision points” and “Audit-ready reports, signed by certified pentesters”.

Terra’s platform uses several AI agents, each handling a different stage of an attack, with human security experts overseeing them, Calcalist reported.

Why Companies Pay for It

The card industry’s security standard, PCI DSS v4.0, requires internal and external penetration testing “At least once every 12 months” and “After any significant infrastructure or application upgrade or change”, per the requirement text quoted on Microsoft’s PCI compliance page.

The same requirement says weaknesses found must be corrected and “Penetration testing is repeated to verify the corrections.” Service providers that isolate card data must test that isolation “At least once every six months”.

That makes at least two penetration tests a year for a company in scope, one internal and one external, plus retests after big changes, by our reading.

The Deal Economics

Terra was founded in 2024 and has raised $38 million: an $8 million seed round in April 2025 and a $30 million Series A led by Felicis that September, according to the company.

A $500 million price would be about 13 times what Terra raised, by our arithmetic. No deal has been announced, and the figure is Calcalist’s.

Palo Alto has already been buying. It agreed on 18 November to buy Applied Compute for about $4.5 billion in stock, and earlier this year bought Chronosphere, Console, Embrace and Koi, deals that together total about $8.5 billion with Applied Compute, by our arithmetic from filings TechCrunch reported.

The Limits

NIST warns that penetration testing “is labor-intensive and requires great expertise to minimize the risk to targeted systems.”

It also says systems “may be damaged or otherwise rendered inoperable” during testing, a risk that “can never be fully eliminated”. Testing “should be performed only after careful consideration, notification, and planning.”

The Structural Read

As we read it, the PCI DSS calendar turns penetration testing into recurring spend, which is the opening continuous AI testing is pitched at.

As we read the vendor pages, the human sign-off is part of the product: certified testers still sign the reports that auditors accept.

NIST Special Publication 800-115

“Penetration testing is security testing in which assessors mimic real-world attacks to identify methods for circumventing the security features of an application, system, or network.”

Three Implications

FOR BUYERS PCI DSS v4.0 requires internal and external penetration testing at least once every 12 months and after significant changes, per the quoted requirement text.

FOR VENDORS Terra claims testing in 2-4 hours instead of 4-6 weeks, and keeps certified pentesters signing reports, per its homepage.

FOR PALO ALTO Calcalist reported talks to buy Terra at a price that could exceed $500 million, about 13 times the $38 million Terra raised by our arithmetic.

What Is Not Established

No deal between Palo Alto and Terra has been announced; the over-$500 million figure is Calcalist’s, from the talks. Terra’s 2-4 hour speed claim is the company’s own, and neither Terra nor OpenAI publishes a price for its AI testing on the pages cited here.

This Week In The AI Supercycle

A World Without an Order, and an AI Central Bank Without a Successor

This week’s issue: the week AI memory stayed largely sold out while the people holding the shares paid: why the AI Supercycle is not a dot-com replay, read through geopolitics, macro and technology, with the gap between building and financing the widest we’ve recorded.

Read the full issue →

Business Engineer Framework

The AI Supercycle

The Business Engineer map of who builds, finances and captures value in AI.

Read the Map of AI →

The Bottom Line

AI penetration testing uses software agents to run hired attacks on a company’s own systems; Terra says certified pentesters sign its reports. Card-industry rules require at least two such tests a year for companies in scope, and Palo Alto may pay over $500 million, about 13 times what Terra raised, for one of the makers, Calcalist reported.

97,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.

Frequently Asked Questions

What is a penetration test?

NIST Special Publication 800-115 defines it as security testing in which assessors mimic real-world attacks to identify methods for circumventing the security features of an application, system, or network. It often involves launching real attacks on real systems and data.

What is AI penetration testing?

Software agents run much of the testing. Terra says its platform unifies continuous agentic pentesting across AI systems, network infrastructure and web apps, with human-on-the-loop at critical decision points and reports signed by certified pentesters.

How often do companies have to run penetration tests?

Under PCI DSS v4.0, as quoted on Microsoft's compliance page, internal and external penetration testing must each be performed at least once every 12 months and after any significant infrastructure or application upgrade or change, and testing is repeated to verify corrections.

Why is Palo Alto Networks interested in Terra Security?

Calcalist reported on 22 Nov 2026 that Palo Alto is in talks to buy Terra in a deal that could exceed $500 million. Terra has raised $38 million, so that would be about 13 times what it raised, by our arithmetic. No deal has been announced.

A note on sourcing. The definition and the limits are from NIST Special Publication 800-115. The PCI DSS v4.0 testing rules are the requirement text as quoted on Microsoft’s PCI DSS compliance page. Terra’s descriptions and speed claim are from its homepage, and OpenAI’s from its Daybreak product pages as quoted in our earlier piece. The Terra talks and price are as reported by Calcalist, and the funding as stated by Terra and reported by SecurityWeek. The 13 times multiple and the two-tests-a-year reading are ours.

Sources: NIST Special Publication 800-115, Technical Guide to Information Security Testing and Assessment (September 2008) · PCI DSS v4.0 Requirement 11.4 text as quoted by Microsoft Learn, Microsoft Entra ID and PCI-DSS Requirement 11 · Terra Security homepage (read 22 Nov 2026) · Our piece on OpenAI’s Daybreak Red Team pen-testing agents (12 Nov 2026), quoting OpenAI’s product pages · Our piece: Palo Alto’s Terra Deal Could Top $500M (22 Nov 2026), citing Calcalist, Terra and SecurityWeek, and filings via TechCrunch · Our piece: 50+ Leaders Expect 18 Months of Constant Cyberattacks (22 Nov 2026), citing Axios

Scroll to Top
FourWeekMBA