CVE-2026-57112: PraisonAI: DNS rebinding exposes MCP SSE tool calls

HIGHPoC AVAILABLECISA: ATTEND

Published September 15, 2026

CISO Take

PraisonAI's legacy SSE MCP server (ToolsMCPServer.run_sse) mounts /sse and /messages/ with no authentication or Host/Origin validation, so a malicious web page can use DNS rebinding to reach a local or internal instance and enumerate and invoke every registered tool with the server user's privileges. Exploitation is hard and needs a user to visit an attacker page (CVSS 8.3, AC:H/UI:R), EPSS is low at 0.00224 (top 88th percentile), and there is no KEV listing or public exploit, but CISA SSVC rates it ATTEND. The risk is concentrated on developer workstations and internal hosts running PraisonAI MCP servers whose tools can touch files, shells, or credentials. Upgrade praisonaiagents to 1.6.59+ and PraisonAI to 4.6.59+, and note that the vendor describes this as an initial remediation, so keep tracking the advisory GHSA-vmf9-xx9w-86wx. Until patched, switch to the Streamable HTTP transport, which rejects hostile Origins, or stop exposing the SSE endpoint.

Sources:NVDEPSSGitHub AdvisoryATLASCISA SSVC

What is the risk?

Medium-to-high. Impact is high (confidentiality, integrity and availability, with scope change because tool execution runs with the server user's privileges), but exploitation needs DNS rebinding, a reachable local or internal SSE server, and a user visiting a hostile page. EPSS 0.00224 and no public exploit or Nuclei template suggest low near-term mass exploitation. SSVC ATTEND and the 'initial remediation' wording mean the fix may be incomplete. Exposure is greatest for developer machines and internal agent hosts, and there is no package-level dependent data (Unknown).

How does the attack unfold?

Initial Access

Victim visits an attacker-controlled website that uses DNS rebinding to reach the local or internal PraisonAI SSE server.

AML.T0078

Discovery

Script sends requests with attacker-controlled Host and Origin headers to /sse and enumerates all registered tools.

AML.T0084.001

Execution

Attacker posts tool calls to /messages/, running tools with the server user's privileges.

AML.T0053

Impact

Sensitive data is read and exfiltrated or systems are modified through the invoked tools.

AML.T0086

Initial Access

Victim visits an attacker-controlled website that uses DNS rebinding to reach the local or internal PraisonAI SSE server.

AML.T0078

Discovery

Script sends requests with attacker-controlled Host and Origin headers to /sse and enumerates all registered tools.

AML.T0084.001

Execution

Attacker posts tool calls to /messages/, running tools with the server user's privileges.

AML.T0053

Impact

Sensitive data is read and exfiltrated or systems are modified through the invoked tools.

AML.T0086

What systems are affected?

Package	Ecosystem	Vulnerable Range	Patched
PraisonAI	pip	—	No patch
1 dependents61% patched~23d to patchFull package profile →
PraisonAI Agents	pip	—	No patch
20 dependents63% patched~21d to patchFull package profile →

How severe is it?

CVSS 3.1

8.3 / 10

EPSS

0.2%

chance of exploitation in 30 days

Higher than 13% of all CVEs

Source: EPSS v3 — FIRST.org

Exploitation Status

Exploit Available

Exploitation: MEDIUM

Sophistication

Moderate

Exploitation Confidence

medium

○CISA SSVC: Public PoC

○Public PoC indexed (trickest/cve)

Composite signal derived from CISA KEV, VulnCheck KEV, CISA SSVC, EPSS, Metasploit, Exploit-DB, trickest/cve, Nuclei templates, and inthewild.io exploitation reports.

What is the attack surface?

AVNetwork

ACHigh

PRNone

UIRequired

SChanged

CHigh

IHigh

AHigh

What should I do?

1 step

1) Upgrade praisonaiagents to 1.6.59 or later and PraisonAI to 4.6.59 or later, and monitor the GHSA for follow-up fixes since the remediation is described as initial. 2) Move to the Streamable HTTP transport, which enforces Origin checks. 3) Bind MCP servers to loopback or a protected interface, place them behind a reverse proxy that enforces an allowed-Host list, Origin validation, and authentication (token or mTLS). 4) Restrict the tools registered on the server to least privilege and run the process under a low-privilege, sandboxed account. 5) Detection: log and alert on requests to /sse and /messages/ with unexpected Host or Origin headers (non-localhost names resolving to private IPs), unexpected tool listing or invocation, and DNS responses for external domains that resolve to RFC1918 or loopback addresses (enable DNS rebinding protection on resolvers).

What does CISA's SSVC say?

DecisionAttend

Exploitationpoc

AutomatableNo

Technical Impacttotal

Source: CISA Vulnrichment (SSVC v2.0). Decision based on the CISA Coordinator decision tree.

How is it classified?

Auth BypassCode ExecutionData LeakageAgentFrameworkPluginAML.T0049- Exploit Public-Facing Application AML.T0053- AI Agent Tool Invocation AML.T0078- Drive-by Compromise AML.T0084.001- Tool Definitions

Which compliance frameworks are affected?

This CVE is relevant to:

EU AI Act

Art. 15-Accuracy, robustness and cybersecurity

ISO 42001

A.6.2.6-AI system operation and monitoring

NIST AI RMF

MANAGE 2.3-Responses to previously unknown risks and mechanisms to supersede or deactivate systems

OWASP LLM Top 10

LLM06-Excessive Agency

Frequently Asked Questions

What is CVE-2026-57112?

Is CVE-2026-57112 actively exploited?

Proof-of-concept exploit code is publicly available for CVE-2026-57112, increasing the risk of exploitation.

How to fix CVE-2026-57112?

1) Upgrade praisonaiagents to 1.6.59 or later and PraisonAI to 4.6.59 or later, and monitor the GHSA for follow-up fixes since the remediation is described as initial. 2) Move to the Streamable HTTP transport, which enforces Origin checks. 3) Bind MCP servers to loopback or a protected interface, place them behind a reverse proxy that enforces an allowed-Host list, Origin validation, and authentication (token or mTLS). 4) Restrict the tools registered on the server to least privilege and run the process under a low-privilege, sandboxed account. 5) Detection: log and alert on requests to /sse and /messages/ with unexpected Host or Origin headers (non-localhost names resolving to private IPs), unexpected tool listing or invocation, and DNS responses for external domains that resolve to RFC1918 or loopback addresses (enable DNS rebinding protection on resolvers).

What systems are affected by CVE-2026-57112?

This vulnerability affects the following AI/ML architecture patterns: agent frameworks, MCP servers, multi-agent systems, developer workstations running local agents.

What is the CVSS score for CVE-2026-57112?

CVE-2026-57112 has a CVSS v3.1 base score of 8.3 (HIGH). The EPSS exploitation probability is 0.22%.

What is the AI security impact?

Affected AI Architectures

agent frameworksMCP serversmulti-agent systemsdeveloper workstations running local agents

MITRE ATLAS Techniques

AML.T0049Exploit Public-Facing Application

AML.T0053AI Agent Tool Invocation

AML.T0078Drive-by Compromise

AML.T0084.001Tool Definitions

Compliance Controls Affected

EU AI Act: Art. 15

ISO 42001: A.6.2.6

NIST AI RMF: MANAGE 2.3

OWASP LLM Top 10: LLM06

What are the technical details?

Original Advisory

PraisonAI is a multi-agent teams system. From praisonaiagents 0.6.0 until 1.6.59 and PraisonAI 3.10.0 until 4.6.59, ToolsMCPServer.run_sse() in src/praisonai-agents/praisonaiagents/mcp/mcp_server.py mounts SseServerTransport on the legacy /sse and /messages/ endpoints without default Host, Origin, or authentication enforcement. A malicious website can use DNS rebinding against a reachable local or internal SSE server, supply attacker-controlled Host and Origin headers, enumerate registered tools, and invoke them with the server user's privileges. The Streamable HTTP transport rejects the same hostile Origin, which isolates the flaw to the legacy SSE wrapper. An initial remediation was released in praisonaiagents 1.6.59 and PraisonAI 4.6.59.

Exploitation Scenario

A developer runs a PraisonAI agent exposing tools via run_sse() on localhost or an internal host. An attacker lures them to a web page, for example through a phishing link. The page's domain first resolves to the attacker's server, then rebinds to 127.0.0.1 or the internal IP, so the browser treats requests to the SSE server as same-origin. Because the server performs no Host, Origin or authentication checks, the attacker's JavaScript opens /sse, enumerates the registered tools, and posts tool calls to /messages/, for example reading files, running commands, or pulling API keys the agent can access, and exfiltrates the results to the attacker.

Weaknesses (CWE)

CWE-306Missing Authentication for Critical FunctionalityPrimary CWE-346Origin Validation ErrorPrimary CWE-862Missing AuthorizationPrimary CWE-306Missing Authentication for Critical Functionality CWE-346Origin Validation Error CWE-862Missing Authorization

CWE-306 — Missing Authentication for Critical Functionality: The product does not perform any authentication for functionality that requires a provable user identity or consumes a significant amount of resources.

[Architecture and Design] Divide the software into anonymous, normal, privileged, and administrative areas. Identify which of these areas require a proven user identity, and use a centralized authentication capability. Identify all potential communication channels, or other means of interaction with the software, to ensure that all channels are appropriately protected, including those channels that are assumed to be accessible only by authorized parties. Developers sometimes perform authentication at the primary channel, but open up a secondary channel that is assumed to be private. For example, a login mechanism may be listening on one network port, but after successful authentication, it may open up a second port where it waits for the connection, but avoids authentication because it assumes that only the authenticated party will connect to the port. In general, if the software or protocol allows a single session or user state to persist across multiple connections or channels, authentication and appropriate
[Architecture and Design] For any security checks that are performed on the client side, ensure that these checks are duplicated on the server side, in order to avoid CWE-602. Attackers can bypass the client-side checks by modifying values after the checks have been performed, or by changing the client to remove the client-side checks entirely. Then, these modified values would be submitted to the server.

Source: MITRE CWE corpus.