November 2026 Summaries

3 posts from Detectify

Filter
Month: Year:
Post Summaries Back to Blog
Detectify’s Agentic Tooling is designed to help security teams automate verification of suspected vulnerabilities, including SAST findings, bug bounty reports, open ports, and internal red-team submissions. Each issue becomes a focused task for an AI agent, which uses Detectify’s testing tools through an MCP server to run assessments against a live application and determine whether the issue is exploitable. The Security Ledger records the assessments, agent actions, and most investigation traffic, providing evidence and traceability for each verdict; findings without supporting tests can be rejected. The product is intended to complement broader DAST scanning by checking specific reported issues rather than continuously searching for any vulnerabilities, and it can work with agents such as Claude Code or OpenCode while customers retain their own models and setups. Detectify says the tooling requires source-code access within the customer’s environment to trace issues, but the company does not access or store the codebase, receiving investigation data and related records instead.
Nov 16, 2026 1,075 words in the original blog post.
Drawing on Doug McIlroy’s Unix philosophy, the passage argues that AI agents should rely on small, focused, composable deterministic tools rather than attempting routine technical tasks through language-model reasoning. It cites research showing that models can produce inconsistent security judgments after superficial code changes and notes reports of high false-positive rates in AI-driven security testing, supporting a division of labor in which tools gather verifiable evidence while models handle contextual decisions. The passage references Anthropic’s guidance on optimizing agent tools, minimizing verbose outputs, and composing operations in code to reduce token use and improve reliability. It recommends exposing high-value workflows through the Model Context Protocol (MCP), using typed inputs, concise responses, clear tool names, and append-only logs that preserve evidence independently of model-generated conclusions.
Nov 12, 2026 1,567 words in the original blog post.
Alpine Linux disclosed CVE-2026-123158, or “sshhole,” on November 5, 2026, an OpenSSH authentication bypass affecting Alpine releases 3.16 through 3.24 and edge when password authentication is enabled and PAM is not used. The flaw occurs because musl’s crypt() returns an asterisk for passwords exceeding 256 bytes, which can match the asterisk used in `/etc/shadow` to represent accounts without passwords, allowing remote attackers to log in to key-only or other passwordless accounts by submitting an overlong password. Official Alpine cloud images can become particularly severe if an attacker accesses the `alpine` account, which may have passwordless `doas` access to root, although root login is normally protected by default settings. Systems are affected only under specific configuration conditions, while hosts using PAM, disabling password authentication, running no SSH daemon, or using non-musl distributions are not vulnerable. Alpine has issued fixed packages, and users are advised to upgrade and restart `sshd` or disable password logins, review logs for suspicious successful password authentications involving accounts with `*` password fields, and treat systems with unverified access as potentially compromised.
Nov 05, 2026 731 words in the original blog post.