Evidence used
- No CISA KEV confirmation is currently recorded.
- The selected CVSS metric records a network-reachable, unauthenticated path with no user interaction.
- EPSS is 0.33% for the current model date.
BlackTreeCVE IntelligenceFull vulnerability report · 2026
CVE-2026-65975High confidence
Pydantic AI AG-UI Adapter: A dangling client-submitted tool call can execute when a trailing message is dropped during `sanitize_messages`
pydantic · pydantic-ai
Official source article: GitHub GHSA-JPR8-2V3G-WGF9 ↗. Check the applicable product and release in the original source.
6.5MediumCVSS 3.1
Patch only the product branches with a verified fix
Fix availability varies by productMedium technical severity with no CISA KEV confirmation; remediate through the normal risk-based patch cycle unless local exposure raises the priority. Verified remediation exists for at least one product or source, but 1 structured product or package state remain unresolved. Apply remediation only to the exact product branch confirmed by its source.
R
Open reassessment dashboard →Operational reassessment
Published severity in operational context
Published severityMedium→Operational priority:Medium, unchanged from published severity.unchanged
Compensating controls
- Apply the linked authoritative mitigation while planning the permanent fix.
- Restrict the affected network interface to trusted sources where business-safe.
- Monitor vendor guidance and exploitation sources for a material change.
Verification
- Confirm that the asset runs pydantic pydantic-ai and falls inside the recorded affected range.
- Recheck the vendor advisory before scheduling a change because no verified fixed version is currently retained.
- Validate exposure, authentication requirements and compensating controls in the actual environment.
- Reopen this reassessment when CVSS, KEV, EPSS, exploit evidence or remediation changes.
Remediation availability differs by product scope
Verified remediation exists for at least one product or source, but 1 structured product or package state remain unresolved. Apply remediation only to the exact product branch confirmed by its source.
- Red Hat Product Security: 1 affected or under-investigation product state without a fixed product state in the same current advisory
Direct vendor intelligence
1 currentAuthoritative vendor CSAF and VEX advisories
Structured product status and remediation from the issuing vendor. Product-state explanations are always visible; large lists can be searched or downloaded.
CVE-2026-65975 · CSAF 2.0 · revision 3 · status not statedRed Hat Product Securitypydantic-ai: Pydantic AI: Remote clients can execute server tools with forged arguments due to improper message sanitization.
The vendor explicitly identifies these products as affected by this CVE.
- lightspeed-core/lightspeed-stack-rhel9 as a component of Lightspeed Core
- Summary
- A flaw was found in Pydantic AI's UI adapters, specifically within the `sanitize_messages` function. This vulnerability allows a remote client to bypass security checks and execute server tools with arguments they provide, rather than those generated by the AI model. This occurs when a trailing client message is removed, causing a previous, unverified tool call to be processed. The impact depends on the functionality of the affected tools, potentially bypassing critical security guardrails.
- Remediation
- Configure Pydantic AI tools to require explicit approval for execution. This prevents remote clients from automatically executing server tools with forged arguments, as approval-gated tools are not affected by this vulnerability. Additionally, restrict network access to the Pydantic AI application to trusted clients only, if feasible in your environment.
Medium technical severity with no CISA KEV confirmation; remediate through the normal risk-based patch cycle unless local exposure raises the priority. Verified remediation exists for at least one product or source, but 1 structured product or package state remain unresolved. Apply remediation only to the exact product branch confirmed by its source.
Fix availability varies by product01
What, why and how
Pydantic AI is a Python agent framework for building applications and workflows with Generative AI. In versions 1.88.0 up to but not including 1.107.1 and 2.0.0b1 up to but not including 2.5.0, the UI adapters (AG-UI via Agent.to_ag_ui()/AGUIAdapter, and Vercel AI via VercelAIAdapter) use sanitize_messages to strip unresolved ("dangling") client-submitted tool calls from untrusted message history before it reaches the agent, a defense-in-depth default that prevents the agent from executing tool calls the model never emitted. However, the strip anchored to a message index computed before sanitization ran, so when a trailing client message sanitized to empty and was dropped (for example a client system message under the default manage_system_prompt='server'), a preceding assistant response carrying an unresolved tool call became the new tail and was dispatched without inspection. As a result, a remote client could cause a registered, non-approval server tool to run with client-supplied arguments rather than arguments the model produced. The impact is bounded by what the affected tools do and is most significant for applications that gate tool execution in a model-request hook (before_model_request / after_model_request), since a forged call skips the model turn and bypasses that guardrail; approval-gated tools (requires_approval=True) are not auto-executed by this path. This issue has been fixed in versions 1.107.1 and 2.5.0.
What
Pydantic AI is a Python agent framework for building applications and workflows with Generative AI. In versions 1.88.0 up to but not including 1.107.1 and 2.0.0b1 up to but not including 2.5.0, the UI adapters (AG-UI via Agent.to_ag_ui()/AGUIAdapter, and Vercel AI via VercelAIAdapter) use sanitize_messages to strip unresolved ("dangling") client-submitted tool calls from untrusted message history before it reaches the agent, a defense-in-depth default that prevents the agent from executing tool calls the model never emitted. However, the strip anchored to a message index computed before sanitization ran, so when a trailing client message sanitized to empty and was dropped (for example a client system message under the default manage_system_prompt='server'), a preceding assistant response carrying an unresolved tool call became the new tail and was dispatched without inspection. As a result, a remote client could cause a registered, non-approval server tool to run with client-supplied arguments rather than arguments the model produced. The impact is bounded by what the affected tools do and is most significant for applications that gate tool execution in a model-request hook (before_model_request / after_model_request), since a forged call skips the model turn and bypasses that guardrail; approval-gated tools (requires_approval=True) are not auto-executed by this path. This issue has been fixed in versions 1.107.1 and 2.5.0.
Why
The application performs an authorisation check, but it does not correctly enforce the required permission boundary.
How
An attacker operating through a network path may attempt exploitation without authentication or user interaction. If successful, the issue may cause the confidentiality, integrity or availability impact described by the vendor.
What
Pydantic AI is a Python agent framework for building applications and workflows with Generative AI. In versions 1.88.0 up to but not including 1.107.1 and 2.0.0b1 up to but not including 2.5.0, the UI adapters (AG-UI via Agent.to_ag_ui()/AGUIAdapter, and Vercel AI via VercelAIAdapter) use sanitize_messages to strip unresolved ("dangling") client-submitted tool calls from untrusted message history before it reaches the agent, a defense-in-depth default that prevents the agent from executing tool calls the model never emitted. However, the strip anchored to a message index computed before sanitization ran, so when a trailing client message sanitized to empty and was dropped (for example a client system message under the default manage_system_prompt='server'), a preceding assistant response carrying an unresolved tool call became the new tail and was dispatched without inspection. As a result, a remote client could cause a registered, non-approval server tool to run with client-supplied arguments rather than arguments the model produced. The impact is bounded by what the affected tools do and is most significant for applications that gate tool execution in a model-request hook (before_model_request / after_model_request), since a forged call skips the model turn and bypasses that guardrail; approval-gated tools (requires_approval=True) are not auto-executed by this path. This issue has been fixed in versions 1.107.1 and 2.5.0.
Why
The application performs an authorisation check, but it does not correctly enforce the required permission boundary.
How
An attacker operating through a network path may attempt exploitation without authentication or user interaction. If successful, the issue may cause the confidentiality, integrity or availability impact described by the vendor.
02
Exploit reality and attack path
CVSS severity, EPSS forecast probability, public exploit material and CISA-confirmed exploitation are separate signals.
Observed exploitation No confirmed evidence
?
Confirmed exploitation and public exploit material are separate signals. Attacks can occur without public proof-of-concept or exploit code.No CISA KEV match was present at the last successful refresh. This means no confirmation from that source, not proof of no exploitation.
Public PoC / exploit material None recorded
?
Confirmed exploitation and public exploit material are separate signals. Attacks can occur without public proof-of-concept or exploit code.No exploit-tagged reference or CISA SSVC proof-of-concept state is currently recorded. Research may still exist outside the structured feeds.
- Likely attack path
- a network path → Incorrect Authorization → cause the confidentiality, integrity or availability impact described by the vendor
- Attack surface
- Network
- Privileges required
- None: unauthenticated exploitation is possible
- User interaction
- None
- Attack complexity
- Low: no specialised conditions are recorded
- Security boundary
- Unchanged: impact remains within the vulnerable component's security authority
- Weakness
?
CWE means Common Weakness Enumeration: a standard category for the underlying weakness. - CWE-863 ↗
CWE-863: Incorrect Authorization. The product performs an authorization check when an actor attempts to access a resource or perform an action, but it does not correctly perform the check.
- CVSS vector
?
CVSS means Common Vulnerability Scoring System. The vector records the metric values used to calculate technical severity. CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:L/I:L/A:NCommon Vulnerability Scoring System 3.1: the compact vector below is decoded into plain language.
AVNetworkAttack vector: The vulnerable component can be reached over a network.ACLowAttack complexity: No specialised conditions are required beyond attacker-controlled input.PRNonePrivileges required: The attacker does not need an account or existing privileges.UINoneUser interaction: No action by another user is required.SUnchangedScope: The security impact remains within the vulnerable component's authority.CLowConfidentiality impact: A successful attack can cause a limited loss.ILowIntegrity impact: A successful attack can cause a limited loss.ANoneAvailability impact: No direct loss is represented by this metric.- Post-exploitation / living off the land
- No specific living-off-the-land technique is confirmed in the structured sources. Monitor normal administration tools for activity inconsistent with the affected service's baseline.
03
Patch and workaround
Operational remediation based on structured source evidence.
- Status
?
Patch availability is based on structured fixed-version fields and authoritative update references. If no fix is verified, check the vendor advisory before making a change. - Fix availability varies by product
- Action
- Use the product-specific evidence above. Patch only products with a verified fixed release, and keep every affected or under-investigation state without a matching fix in the remediation queue.
- Workaround
- Configure Pydantic AI tools to require explicit approval for execution. This prevents remote clients from automatically executing server tools with forged arguments, as approval-gated tools are not affected by this vulnerability. Additionally, restrict network access to the Pydantic AI application to trusted clients only, if feasible in your environment.
04
Evidence and provenance
Published 29 Jul 2026 · Last source change 30 Jul 2026, 13:54 UTC · CWE-863 · Incorrect Authorization
CVE recordCVE.org · 5.2
CVSS sourceCNA
EPSS source FIRST · tracked since 2026-08-14
?
The date BlackTree first stored a score for this CVE from the daily FIRST EPSS feed.European sourceENISA EUVD · EUVD-2026-50555
Product sourceVendor CSAF · Red Hat Product Security
Remediation sourceVendor CSAF · Red Hat Product Security
CWE sourceCNA
NVD statusNVD enriched
Core structured fields are present and their contributing authorities are shown above.
Material change intelligence
View recent updates ↗What changed after publication
No material field changes have been recorded since change tracking began. Routine source refreshes and cosmetic edits are intentionally excluded.
Technical terms and abbreviations used in this report
- CVE
- Common Vulnerabilities and Exposures: the public identifier for one disclosed vulnerability.
- CVSS
- Common Vulnerability Scoring System: a technical severity framework; it is not patching priority by itself.
- EPSS
- Exploit Prediction Scoring System: FIRST's estimate of the probability that exploitation activity will be observed in the next 30 days; it is a forecast, not confirmation.
- CWE
- Common Weakness Enumeration: the standard category describing the underlying software or hardware weakness.
- CNA
- CVE Numbering Authority: an organisation authorised to assign and publish CVE records.
- CISA ADP
- Cybersecurity and Infrastructure Security Agency Authorized Data Publisher: structured enrichment added to a CVE record.
- NVD
- National Vulnerability Database: NIST's enrichment service for CVE records.
- CERT / CSIRT
- A computer security incident response team that publishes warnings or coordinates incident response.
- PoC
- Proof of concept: public material that demonstrates or helps reproduce exploitation.
- CSAF
- Common Security Advisory Framework: a machine-readable format for security advisories.
- LoTL
- Living off the land: abuse of legitimate tools or system functions during an attack.