Pentera told staff on August 17 that roughly 60 more people are out, about 30 of them in Israel. On April 27 the same company cut about 40. Put together, that is close to 100 people in 112 days, or about 20% of a 470-person company.
The headline number is the 20%. The load-bearing number is the 112 days.
Both statements name AI, four months apart
April’s statement was corporate but not coy. Pentera said it was refining structure and resource allocation around “the continued development of its AI-driven cybersecurity verification products,” and that it would keep recruiting across product, engineering and AI innovation. The cuts landed on marketing teams and headquarters roles. Under 10% of the company.
August’s statement moved a step forward. Pentera tied the round directly to the pace at which AI is being developed and adopted, said it is completing the transformation of its platform to make it AI-native, and repeated that AI hiring continues.
Same reason, same action, more explicit language on the second run. Companies rarely repeat a rationale unless they intend to repeat the action.
This is not a company running out of money
Pentera raised $150M in January 2022 at a valuation above $1B. It added roughly $60M in a March 2025 Series D led by Evolution Equity Partners, bringing total capital raised to about $250M. Annual recurring revenue has crossed $100M.
In the same stretch it bought two companies: DevOcean, an all-Israeli deal reported at about $30M, and EVA Information Security. It opened offices in Colorado and Madrid on top of Denver, Boston, Germany, the UAE and Singapore.
A firm past $100M ARR, two acquisitions in, still opening offices, cut a fifth of its people in one third of a year. That is not the shape of retrenchment.
The denominator never moved
Here is the detail worth stopping on. Calcalist ran 470 as the base both times: April’s 40 was under 10% of 470, August’s cumulative 100 is 20% of 470.
If April’s 40 had been a net reduction, the August base would be 430, the new round would compute to 14%, and the cumulative math would need a different denominator. It didn’t. The base held.
Which means the four months between the two rounds were not a period of shrinking. They were a period of swapping. Marketing and headquarters walked out; product, engineering and AI walked in. The company is roughly the size it was in April and is not remotely made of the same people.
Public companies scrub AI from the filing. Private ones say it out loud.
On August 10, Rapid7 raised its profit outlook and cut 12% in the same package. Same sector, same profitable-cut structure, and the 8-K never uses the word AI. The reason is documented: Rackspace wrote its AI pivot into a June 8-K and by August that filing was Exhibit A in a securities class action. Legal departments adjusted within a quarter. The AI story goes on the earnings call; the filing says “simplify operations.”
Pentera has no such constraint. It is private. There is no shareholder class to certify, no forward-looking statement to be measured against, no plaintiff’s exhibit to worry about. So it named AI in April, named it again in August, and named it harder the second time.
That produces a counterintuitive research rule for anyone trying to measure how fast AI is actually replacing roles: private-company announcements are now the cleaner sample. Public 8-Ks have been through counsel. Private press statements have not. The most honest disclosure in this cycle is coming from the companies with the least disclosure obligation.
Add the cohort. On June 16, Artlist cut 40% at record revenue and called it a transition to an “AI-native operating model”. Israeli, profitable, growing, same two words. On August 4, Zillow cut more than 500 and told reporters AI had nothing to do with it, three months after the CEO told investors Zillow was becoming AI-native. Artlist took it in one cut. Zillow took it and denied it. Pentera is running it in installments and saying so.
Which layer is being swapped
April named the functions: marketing teams and headquarters roles. August did not publish a functional breakdown, but the rationale and the hiring posture are identical across both rounds, which points at the same layer.
Concretely that layer is content marketing, demand generation, marketing operations, sales enablement, and the HR, finance and administrative support seats at headquarters. What those roles ship each day is copy, campaign assets, industry reports, lead-scoring sheets, weekly decks and dashboards. Every one of those outputs has a template, a large back catalogue of prior examples, and a tolerance for imperfection that engineering code does not have. That is the exact profile generative models clear first.
Security-validation vendors make the pattern sharper than most. A large share of the marketing function at a company like Pentera exists to translate what the offensive-security engineers produced into something a CISO buyer can read. Translation between a technical artifact and a business artifact, against a known template, is the single task current models handle best.
What a candidate should read from this
The old signal was binary. A company is cutting, so skip it. A company is hiring, so apply. Pentera runs both at once, in the same quarter, and the two facts do not contradict each other. Its careers page and its layoff coverage are two sides of one document.
So the question changes shape. Not “is this company cutting,” but “which layer is it cutting and which layer is it filling.” For an AI or platform engineer, Pentera in August 2026 is an expanding employer. For a marketing operations manager, the same company on the same day is a closing door. One employer, two opposite conclusions, and no way to tell them apart from the headline.
There is a clean thing to watch, with a date on it. The gap between round one and round two was four months. If a third round lands before year end, this is a quarterly cadence rather than a one-time restructure, and everyone holding a go-to-market or back-office seat in security software should plan against a structural 20% a year rather than a bad twelve months. If year end passes quietly, April plus August was a single rebuild that is now finished, and the floor holds.
That answer arrives inside the next 130 days.
