/

CVE-2026-92576

Severity:

HIGH RISK

(8.6)

SSRF (CWE-918) in HKUDS nanobot WebFetchTool Component

CVE-2026-92576

Deferred

Vulnerability type

Category:

Server-Side Request Forgery

CWE:

CWE-918

Impact:

Unauthorized Access to Internal Services and Credential Exposure

Severity:

HIGH RISK

(8.6)

CVSS version:

3.1

Catch this CVE while code is being written

Detect vulnerable code directly in pull requests, so CVEs are fixed before merge.

Summary

HKUDS nanobot contains a server-side request forgery vulnerability in its WebFetchTool component because URL validation does not adequately block private and internal destinations.

An attacker who can send instructions to the bot may cause it to access cloud metadata services, localhost endpoints, or RFC 1918 addresses, potentially exposing credentials and internal service data.

Affected context

The issue occurs in deployments that expose nanobot's message-driven WebFetchTool and allow attacker-controlled or insufficiently trusted instructions to influence outbound URL requests. Risk is higher when the bot runs in a cloud or internal network with access to metadata endpoints, administrative services, or other sensitive resources.

Exploit preconditions

  • Attack surface: The WebFetchTool's outbound HTTP request functionality
  • Required access: Ability to send messages or otherwise provide instructions processed by the bot
  • Trigger point: A request that causes the bot to fetch an attacker-selected URL
  • Why exploitation is possible: URL validation does not reliably reject internal IP ranges, private addresses, localhost services, or cloud metadata destinations

Why this happens

The vulnerability results from incomplete server-side URL validation. The application treats user-influenced destinations as fetchable without enforcing a robust network-boundary policy, allowing requests to resolve to internal or otherwise restricted resources.

This behavior maps to CWE-918: Server-Side Request Forgery, where a server performs network requests on behalf of an attacker and can reach destinations that should not be externally controllable.

Potential impact

  • An attacker may retrieve cloud instance metadata
  • IAM credentials or temporary access tokens could be exposed
  • Internal service data may be disclosed
  • Localhost administrative interfaces could be accessed
  • Network reconnaissance and lateral movement may become possible
  • Sensitive responses could be returned through the bot

General mitigation guidance

Mitigation generally focuses on reducing trust in user-supplied URLs and enforcing outbound network restrictions.

Common defensive patterns include:

  • Permit only explicitly approved URL schemes, hosts, and destinations
  • Resolve hostnames and validate every resulting address against private, loopback, link-local, multicast, and reserved ranges
  • Revalidate destinations after redirects and prevent DNS-rebinding bypasses
  • Apply egress firewall policies that deny access to metadata services and sensitive internal networks
  • Use isolated, least-privileged execution environments for fetch operations
  • Avoid returning unrestricted upstream responses to untrusted users

How teams detect issues like this

Organizations can detect SSRF exposure through:

  • Static analysis for server-side HTTP clients receiving user-controlled URLs
  • Code review checks for incomplete private-address and redirect validation
  • Dynamic testing with loopback, RFC 1918, link-local, metadata, and DNS-rebinding cases
  • Network monitoring for unexpected requests from bot workloads to internal services or cloud metadata endpoints
  • Policy enforcement requiring allowlists and controlled egress for automated fetch components
  • Automated security review tools that flag unvalidated outbound request destinations

Sources

SourceLink
github.comgithub.com/HKUDS/nanobot/security/advisories/GHSA-vc5v-6vwm-wf9m
vulncheck.comvulncheck.com/advisories/hkuds-nanobot-before-0.3.0-server-side-request-forgery-via-webfetchtool

Frequently asked

questions

Potentially. If an attacker can send messages that control WebFetchTool requests, exploitation may occur remotely through the bot's exposed interaction channel.

The attacker generally needs the ability to submit instructions processed by nanobot. The required level of access depends on authentication and authorization controls around the bot.

Use an allowlist-based fetch policy, block private and special-purpose address ranges after DNS resolution, validate redirect targets, and enforce network-level egress restrictions.

Treat all requested URLs as untrusted, combine application-layer validation with firewall controls, isolate fetch workers, and continuously test for internal-network access paths.

Find out what's already
exploitable in your codebase.