Summary
AutoGPT's workflow platform contains a server-side request forgery (SSRF) issue that lets authenticated users bypass private-IP protections in the SendWebRequestBlock component.
By abusing IPv4-mapped IPv6 hostnames and unblocked special-use ranges, an attacker can make the backend reach internal network services that should be isolated.
Affected context
The vulnerability occurs in AutoGPT Platform deployments where backend workflows allow authenticated users to configure or trigger outbound HTTP requests via SendWebRequestBlock.
It is present when IP blocking logic relies on direct comparisons against IPv4 ranges without normalizing IPv4-mapped IPv6 addresses or considering certain special-use ranges such as 100.64.0.0/10.
Under these conditions, hostnames that resolve to IPv4-mapped IPv6 addresses or CGNAT ranges may slip past filters and be reachable from the backend environment.
Exploit preconditions
- Attack surface: The SendWebRequestBlock functionality exposed through authenticated workflow or agent configuration interfaces.
- Required access: Valid authenticated access to the AutoGPT Platform with permission to create, edit, or trigger workflows that issue HTTP requests.
- Trigger point: Supplying a crafted hostname or URL that resolves to an IPv4-mapped IPv6 address or an unblocked special-use range, causing the backend to perform the request.
- Why exploitation is possible: The IP blocking routine
_is_ip_blocked()does not normalize IPv4-mapped IPv6 addresses before evaluating them against IPv4 blocklists and omits certain special-use ranges from its checks, allowing internal or semi-internal endpoints to be reached.
Why this happens
The root cause aligns with CWE-918: Server-Side Request Forgery, where a server component performs attacker-controlled outbound requests without correctly enforcing network boundaries.
In this case, the backend relies on IP-based blocking to prevent access to private or internal ranges, but the implementation only evaluates raw resolved addresses against IPv4 CIDR filters.
Because IPv4-mapped IPv6 addresses are not normalized to canonical IPv4 form before checking, and certain special-use ranges (e.g., 100.64.0.0/10) are not included in the blocklist, the validation logic incorrectly treats these destinations as allowed.
This pattern reflects a broader class of SSRF defenses that focus on incomplete IP checks instead of robust, context-aware network and application-level controls.
Potential impact
- Authenticated users may cause the backend to send HTTP requests to internal services not intended to be exposed.
- Attackers could probe internal network topology and service behavior from the AutoGPT environment.
- Sensitive metadata or configuration data accessible via internal HTTP endpoints can be exposed.
- Internal APIs or management interfaces could be abused, depending on what is reachable and how they are secured.
- SSRF can potentially be chained with other weaknesses to escalate privileges or gain broader system compromise.
General mitigation guidance
Mitigation should focus on hardening outbound request handling and reducing trust in user-supplied destinations.
Useful patterns include:
- Performing strict normalization of all resolved IP addresses (including IPv4-mapped IPv6) before applying blocklists and range checks.
- Treating access to any non-public or special-use network ranges as denied by default, and only allowlisted under carefully controlled conditions.
- Applying higher-level constraints on where automated agents and workflow blocks may send HTTP requests (e.g., only specific domains or services).
- Isolating components that perform outbound requests in tightly controlled network segments with limited egress.
- Ensuring that internal management and metadata services are not reachable from generic HTTP client code paths, even if SSRF occurs.
How teams detect issues like this
Detection usually combines code-centric and operational approaches:
- Static analysis or manual code review to identify custom IP-blocking or SSRF protection logic, checking for normalization of IPv4-mapped IPv6 addresses and coverage of special-use ranges.
- Verification that hostname resolution and IP comparison routines use well-tested libraries that handle IPv4/IPv6 edge cases consistently.
- Security tests or dynamic analysis that attempt requests to private, loopback, link-local, CGNAT, and IPv4-mapped IPv6 targets to validate SSRF defenses.
- Policy-based checks in CI/CD to flag introduction of new outbound HTTP capabilities lacking centralized SSRF protection wrappers.
- Monitoring backend logs and network telemetry for unusual outbound requests to internal ranges or unexpected destinations initiated by user workflows.

