Record-breaking autonomous attack security company extends full-stack testing to include AI systems, covering web, API, and AI application layers in a single run.
New York City, NY / Access Newswire / June 1, 2026 / Sugar beetThe autonomous attack security company, which ranks in the top 1% of 125,000 human hackers across six global platforms, today announced that it is expanding its AI hackers to include AI applications. This enables enterprise security teams to test the complete attack surface of modern AI systems, from the web layer and APIs to the AI applications themselves.
Today’s announcement comes as AI applications enter production at a faster pace than existing security infrastructure. gartner project 40% of enterprise applications will include task-specific AI agents by the end of 2026, up from less than 5% in 2025. A significant portion of these applications are built by AI coding agents. Tenzai’s original research We found that all major AI coding tools (Cursor, Claude Code, Codex, Devin) ship vulnerable code when tested against the same prompts and environments. Independent research supports this pattern. 45% of AI-generated code samples introduce OWASP Top 10 vulnerabilitiesAnd the results are even worse – In March 2026 alone, more CVEs were directly attributable to AI-generated code than in all of 2025 combined.
A different approach to AI application security
Tenzai’s approach to AI application security differs from market standards. The majority of AI application security tools on the market today are built from outside the model. That is, test whether the model resists immediate injections, whether training data is leaked, and whether guardrails are held. Tenzai’s approach starts across the offensive line and works its way inside.
This distinction is important when considering what vulnerabilities in AI applications actually look like. Prompt injection is often the entry point, but the underlying vulnerability is excessive tool privileges, missing authorization checks, or credentials propagated to workflows that should not be inherited. These are classic security bugs made reachable through the AI layer. Finding them requires an agent that understands both the behavior of the AI and the infrastructure on which it operates.
When Tenzai’s agent encounters an AI application, the prompt insertion checklist is not run against the model. Map your application as a set of actors, instructions, tools, credentials, guardrails, state transitions, and endpoints, and generate leads across them all.
Pavel Gurvich, CEO and co-founder of Tenzai, said, “The most critical vulnerabilities in AI applications are not in the models, but in what model-driven agents have access to.” “Extending Tenzai’s AI hackers to AI applications is a natural next step. Our hackers are already continuously tracking the behavior of the web layer, API, and now AI in the same run, and every test builds on what was learned in the last. Important discoveries are almost always cascaded, so if you test one layer in isolation or once a quarter, you’ll miss them.”
Watch a demo and read more about Tenzai’s AI application hacking here. link
About beetroot:
Tenzai is an AI-native cybersecurity company that builds autonomous AI hackers to help enterprises reliably deliver unbreakable codes. Its platform helps proactively hack, exploit, and remediate vulnerabilities across enterprise software, continuously and at scale. Founded in 2025 by cybersecurity veterans Pavel Gurvich, Ariel Zeitlin, Ofri Ziv, Itamar Tal, and Aner Mazur, Tenzai has raised $75 million in seed funding from leading investors including Battery Ventures, Greylock Partners, Lux Capital, and Swish Ventures. read more: www.tenzai.com
Media inquiries:
Itai Singer, TellNY
Itai@tellny.com
sauce: Sugar beet
Read the original press release on ACCESS Newswire
