AI has changed everything. Are you ready?
Attackers now use AI to find weaknesses and turn them into working attacks in hours instead of months. On November 16, 2026, Google reported, for the first time, attackers using AI to find and exploit software flaws with no human in the loop[1]. Would you find those weaknesses before they do? We can help.
Penetration testing for SOC 2, PCI DSS, HIPAA, ISO 27001, and customer security reviews.
What changed in 2026.
The gap between a published flaw and a working attack used to be measured in weeks or months. In 2026 it collapsed.
Google found attackers using an AI framework that compares code before and after a patch, writes the exploit, and runs it without human intervention.[1]
Given only a public patch, an AI model wrote its first working Firefox exploit in under an hour.[2]
Four researchers and a publicly available AI model built a full Google Chrome exploit chain in six days. Teams once spent about six months on that kind of work.[3]
Published vulnerabilities and exploits nearly doubled, and zero-day exploits jumped to 22 in August, against a 2025 average of 8 a month.[1]
Those same tools are aimed at companies like yours right now. If an attacker finds the way in first, you hear about it from a ransom note, a reporter’s call, or the breach notice you have to send every customer. We find it first, using the same techniques attackers use, AI included, and help you fix it before your company becomes the next headline.
Can AI test you on its own? Not yet.
AI covers ground fast. It also gets things wrong, and the people who use it every day know it.
would trust AI alone to chain an exploit
In a September 2026 survey, only 2% of penetration testers would let current AI chain a known exploit into a multi-step attack without supervision, even though 61% used AI frequently or extensively.[4]
run AI agents at full autonomy
In a 2026 survey of security leaders, only 7% ran AI agents at full autonomy. The rest kept a person involved.[4]
say a person must check the results
The share of testers who say a person must validate and interpret automated results rose from 16% in January 2026 to 37% in September 2026.[4]
Even attackers’ AI makes things up. In the espionage campaign Anthropic disrupted in 2025, the AI invented credentials and overstated what it had stolen[5].
So we use AI for reach and people for judgment. An expert tester directs the work, proves each finding by hand, and stands behind your report.
You decide how much AI touches your systems.
Some clients prohibit AI entirely. Others welcome frontier models. We test at whatever level you are comfortable with, and we confirm the level with you before testing begins.
No AI
Fully manual testing and reporting. Nothing about your environment is processed by an AI system.
Private and zero-retention models
Models we control, or commercial models under agreements where the provider keeps nothing and trains on nothing.
Frontier models
Leading commercial models, used under our direction, for more coverage in the same window.
A person is in the loop at every level.
Most clients call us before an attack. Some call after.
Before, we find the weaknesses an attacker would use and help you fix them. After, we find out how the attacker got in, what they reached, and how to close the door behind them. An ounce of prevention is still worth a pound of cure.
Penetration Testing
Applications, APIs, networks, and cloud, tested from outside, from inside, or with your source code.
Learn more →AI / LLM Security Testing
Your chatbots, agents, and AI features, tested against the attacks built to break them.
Learn more →Red Team & Social Engineering
Full-scope attacks on your people, processes, and technology, from a phishing email to a full adversary simulation.
Learn more →IoT, Embedded & Hardware
Two decades of breaking hardware, from medical devices to consumer routers. More than 200 million devices and counting.
Learn more →Testing Programs
Testing on the schedule you choose, with attack surface monitoring in between, so new exposure turns up in days.
Learn more →Incident Response
Call us before an attack to find the way in, or after one to find out how they got in.
Learn more →Your auditor needs a test. You need to know what an attacker could do.
Most clients come to us because a standard calls for a penetration test, most often SOC 2, but also PCI DSS, HIPAA, ISO 27001, an insurer, or a customer’s security review. You get a report your auditor can use, and findings your team can act on, because we test the way an attacker would.
Audit deadline coming up? We can often start within 24 hours.
Five clean reports. One finding that reached every customer’s data.
Real engagement, generalized at the client’s request.
A client engaged us to test a multi-tenant application during acquisition diligence. The target had five clean annual pentest reports, and the critical assurance was that each customer’s data lived in physically separate data stores. Cross-tenant access, and access to production data, was supposed to be technically impossible.
We tested the data boundary behind that claim. When an unexplained session value was removed, the application still worked without error, but the records returned quietly changed. Followed to the end, that behavior crossed tenant boundaries and reached production, exposing every customer’s records.
See more cases, and how we work →- Info An unexplained session value can be removed without breaking the page
- Low The application still returns normal-looking results
- Med The records returned change across tenant and environment boundaries
- Critical Production data from every customer at every site becomes reachable
Trusted where it counts.
A clear process, every engagement.
Scope
We define objectives, targets, and rules of engagement around your real risk, not a template.
Test
We test the way an attacker would, using custom tooling and only the level of AI you approve, and chain real vulnerabilities into proven, exploitable impact.
Report
Clear, prioritized findings with proof-of-concept evidence and remediation your team can act on.
Retest
We validate your fixes and confirm the risk is closed. Included, not upsold.
Find out where you stand before an attacker does.
A penetration test costs a fraction of a breach. Spend a few days finding the flaw, not millions recovering from it.
Not ready to scope a test? Just ask a question. Every message reaches an expert directly and gets a personal reply, not a sales call.
Sources
- Google Threat Intelligence Group, GTIG AI Threat Tracker, Agentic Threat Activity Crosses a Significant Threshold, November 16, 2026.
- SecurityWeek, Claude Mythos Turns N-Days Into N-Hours, June 9, 2026. Reporting Anthropic research.
- XBOW, XBOW Built a Chrome Full-Chain Exploit, October 13, 2026. Built August 2 to 7, 2026.
- Cobalt, 5 Takeaways From the Autonomous Pentesting Pulse Report, November 17, 2026. Tester figures from surveys of Cobalt’s own tester community (198 in January 2026, 94 in September 2026). The 7% figure is from an Omdia survey of 400 North American IT and security professionals, commissioned by Cobalt.
- Anthropic, Disrupting the First Reported AI-Orchestrated Cyber Espionage Campaign, November 13, 2025. Anthropic’s own account.