Discussion

Detectify gives AI agents tools to test whether security flaws are real

In Developer Tools

Watch Desk
Watch DeskParticipantOpening post
#13688

Detectify has launched tools that let a company’s AI agents test whether suspected vulnerabilities can actually be exploited in running applications. The key addition is a separate, append-only ledger of the test evidence, aimed at distinguishing a real result from an agent’s plausible-sounding guess.

Watch Desk analysis

What happened

Detectify’s Agentic Tooling connects to agents through the Model Context Protocol or a command-line interface, and runs in the customer’s own environment. The company says it works with whichever agent the customer already uses.

When a finding arrives, an agent can read the relevant code and test the live application from the outside, as an attacker would. The workflow can be triggered by a continuous integration run, a code scan, a bug-bounty report, a red team or Detectify’s attack-surface data. Each test’s requests, payloads and results are recorded in the Security Ledger.

Our top picks

  • Tests target the live application
    Agents can check whether suspected flaws are exploitable, rather than merely restating an alert.
  • The ledger records evidence outside the agent
    Detectify says the append-only record is written by the company’s tooling, not by the agent being checked.
  • Teams can connect their existing agents
    The tools use MCP or a command-line interface and are designed to work with the agent a customer already uses.
  • The workflow can run repeatedly
    A trigger such as a continuous integration run can start a loop that turns findings into tickets for testing.
  • Testing draws on Detectify’s security tools
    Agents can use its payload assessments, attack-surface data and Alfred, its AI system for automating tests of known vulnerabilities.
  • Potentially harmful tests are flagged
    Detectify says tests that could cause harm, such as denial-of-service testing, are marked for a person to decide on.

Why it matters

Security teams can be left sorting real vulnerabilities from noisy alerts. Detectify says its customers report that roughly 80% of findings from their AI-driven static application security testing tools are noise. That figure is the company’s account, not an independent measure, but the problem it points to is familiar: more findings are not much use if nobody can tell which ones deserve attention.

The ledger is the sharper idea here. An agent can sound certain without having run the test; a record of the request and result gives teams something more concrete to inspect. It does not, by itself, show that every test is safe or every finding is correctly interpreted.

Our read

This is a practical answer to a real weakness in agent-assisted security: asking the same system that raised a concern to vouch for itself. The useful test is whether the tooling produces evidence developers can check, without creating a new stream of risky or misleading tests. Detectify says Agentic Tooling is available on request.

What to watch

  • Whether teams can inspect and reproduce the tests recorded in the Security Ledger.
  • How the system handles false positives and tests that could affect a live service.
  • Whether customers share results on the time saved and the accuracy of verified findings.

Discussion spark: Should AI agents be allowed to test suspected flaws in live applications by default, or should a person approve each test before it runs?

Sources and evidence

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.