Pentera Opens Early Access to Agentic Web App and API Pentesting
Pentera moved its agentic web and API pentesting product into early availability on November 18, 2026, after a beta with more than 60 customers. General availability now slips to Q1 2027.

Why it matters
- Pentera announced early availability of Pentera Apps on November 18, 2026, after a beta with more than 60 customers.
- General availability slipped from Q4 2026 to the first quarter of 2027.
- Beta findings included unpaid e-commerce checkouts, MFA bypass, account takeovers and SQL injection in an unauthenticated API.
- A live agent attack on an e-commerce application is planned for December 1, 2026, at 11:00 a.m. ET.
- Model agreements provide zero retention and prohibit training on customer data, Pentera says.
Pentera moved Pentera Apps, its agentic web application and API penetration-testing product, into early availability on November 18, 2026 — and more than 60 customers had already tested it in beta. The launch matters because it puts autonomous AI agents inside one of security's most sensitive jobs: attacking live applications on behalf of the organizations that own them, under controls the company says stop bad tests before a single packet is sent.
The beta produced real findings, not lab exercises. Pentera reported cases involving unpaid e-commerce checkouts, account takeovers achieved through broken access control, multifactor-authentication bypass, and customer-data extraction through SQL injection in an unauthenticated API. Those cases span payment logic, user permissions and API input handling. Together they show the product's intended focus on application behavior as well as the familiar vulnerability classes security teams already hunt.
Why the schedule moved
The release follows the selected-customer beta that Pentera announced on July 29, 2026. At that point, general availability was targeted for the fourth quarter of 2026. The new schedule places it in the first quarter of 2027 — a slip of roughly one quarter for a product that joins Pentera's existing network, cloud and identity validation offerings. The company describes the combined portfolio as the only all-attack-surface exposure-validation platform, a claim competitors will now be measured against.
How does the agentic testing actually work?
The product documentation describes an attack engine that maps application behavior and tests workflows under different user roles and permissions. Specialist agents handle individual attack classes. They adjust payloads as the application responds, and they generate and run exploit code. Confirmed weaknesses can be linked into paths that extend from applications and APIs into internal networks, Active Directory and cloud identity-and-access-management systems — the difference between finding a bug and showing how far an attacker could ride it.
Testing can start from a URL, with no source-code access required, although credentials and documentation can deepen coverage. The system maps routes, hidden parameters, and documented or undocumented APIs. It supports REST, GraphQL and WebSocket interfaces. It also tests authenticated workflows behind single sign-on, multifactor authentication and OAuth — the layers where most modern business logic actually lives. Tests can run on demand, on a schedule, or when endpoints, authentication or permissions change, with incremental retesting limited to what changed rather than a full rescan.
Context comes from what Pentera calls the Validation Grid. It holds asset profiles, validated findings, threat-actor techniques and signals from across the technology stack. Customers can add earlier penetration-test reports and environment documentation. Pentera says this material guides the agents' investigation, while the execution harness determines which actions can run against the target.
What stops the agents from running wild?
This is the question every buyer will ask, and the release answers it in detail. Before execution, a control plane checks targets, authorization, credentials and testing policies. Invalid configurations stop a test before traffic is sent. Pentera says testing can target live production applications using non-destructive payloads within an approved scope, and users can stop a test at any time. Deterministic checks confirm exploitability, and dedicated validator agents review each result; conflicting signals are left inconclusive rather than promoted to findings.
Data handling follows the same pattern of explicit boundaries. Pentera says sensitive information is removed before the context required for testing is sent to AI models. Its enterprise agreements with model providers provide for zero retention and prohibit training on customer data. Test credentials are held in memory during execution and discarded afterward, while customer data is processed and stored in an assigned EU or US region.
The release frames the operating boundary plainly: AI agents can generate and execute attacks, but the customer defines the scope, the harness controls execution and validators determine what becomes a finding.
From exploit evidence to retesting
Findings can include proof of exploitation, request-and-response records, an execution trace, reproduction steps and a replayable script. That replayability is the design choice most likely to separate this product from scanner noise — evidence a human can rerun, not a heuristic score.
Pentera Apps does not repair the weaknesses itself. Instead, Pentera Resolve assigns each finding to the responsible team under the organization's service-level agreement, tracks remediation by that team or through integrated workflows, and triggers another attack-path run to confirm that the exposure is actually closed. The loop from finding to verified fix is the product's real pitch.
The company is also explicit about what it does not replace. The product takes code-analysis and scanner findings as inputs for runtime testing, not as competition. It does not replace manual penetration testing or bug bounties. Its stated role is continuous validation between engagements and as applications change.
A live attack on December 1
Pentera plans a live demonstration on December 1, 2026, at 11:00 a.m. ET, in which an agent will attack an e-commerce application. Planned demonstrations include workflow reasoning, chained weaknesses, refusals at scope boundaries and replayable exploit evidence. One application will be selected for the live test, and organizations that own or are authorized to test a proposed application could submit it through the Call for Apps until November 23, 2026.
For AppSec teams, the stakes are straightforward. Autonomous testing tools have historically traded coverage for credibility, generating alert streams nobody trusts. The practical question this launch raises — and that the December 1 demonstration will start to answer — is whether replayable evidence, scope refusals and automated confirmation retesting can turn continuous autonomous testing into completed fixes rather than another queue of unverified findings.
Source: unite.ai
More from Elena Vasquez
Show full bio
Market editor covering technology and product news in artificial intelligence and machine learning at AI In Context.
708 articles
Related articles
- Novee adds external attack surface testing to its AI pentesting platform
- Google Announces Gemini 4 Argon, but No One Can Use It Yet
- Google’s Ask AdMob and AI Agent Tools Target Ad Workflows
- Google Ships Gemini 4 Argon to Paying Users Only
- Google Ships Gemini 4 Argon, but Wall Street Wants a Consumer Agent