Chainlit 2.12.0 patches critical unauthenticated RCE and SSRF flaws in MCP feature

·1 views

Two security vulnerabilities, CVE-2026-45018 (CVSS 9.8) and CVE-2026-45019 (CVSS 7.2), were disclosed in Chainlit versions 2.4.0rc0 through 2.11.0, affecting installations with the MCP feature enabled. The first flaw allowed unauthenticated attackers to execute arbitrary commands via the stdio transport by supplying a crafted command string that bypassed incomplete argument validation. The second vulnerability enabled unauthenticated server-side request forgery through SSE and streamable-HTTP endpoints, which made outbound requests to internal or metadata hosts using caller-controlled headers. Both issues were reported by SPL Security researchers Vipin and Stephen, who demonstrated working exploits against version 2.11.0. Chainlit 2.12.0, released on August 25, 2026, removes client-supplied command execution and addresses the SSRF path; operators are advised to upgrade immediately and restart the service to ensure the patched code is loaded.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

How US Startups Should Pick European SMS APIs: Prioritize Reliability Over Cost

A new guide advises US startups to evaluate European SMS alert APIs based on the cost of a failed message rather than the cheapest advertised rate. The framework recommends starting with a single transport provider behind an adapter, while requiring route tests, sender registration proof, inbound reply verification, and an application-owned delivery ledger. Developers are warned that a successful API response only confirms acceptance, not delivery, and that accepted and delivered states must be tracked separately. Sender registration and two-way SMS support both require end-to-end handset testing, as dashboard indicators do not reflect what recipients actually see. The guide stresses that 'Europe' is a planning label, not a delivery guarantee, and that each destination country, sender type, and message template needs its own test row before launch.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Apache Hive 4.x Critical Flaw Lets Attackers Impersonate Any User via Forged Token

Apache disclosed CVE-2026-53561 on August 25, 2026, a critical authentication bypass in HiveServer2 affecting Apache Hive versions 4.0.0 through 4.2.0. The vulnerability allows an unauthenticated attacker with network access to the HiveServer2 HTTP endpoint to forge a Bearer token and gain a session as any arbitrary Hive user. The flaw only affects deployments using HTTP transport with SAML authentication enabled, not the default Kerberos or LDAP configurations. Researcher Andrew Rukin of Arenadata discovered and reported the bypass, which is classified under CWE-287 improper authentication. Apache has released version 4.2.1, which patches this issue alongside a Metastore SQL injection flaw and an Avro SerDe SSRF vulnerability.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Three Rules for Separating Health Data Authentication, Consent, and Access Control

Engineers handling health data should treat login, consent grants, and data access as three distinct decisions rather than collapsing them into a single session-based permission. A consent grant must specify the subject, health data category, purpose, and policy version — not act as a blanket boolean approval for all records. Revocation should immediately affect future reads and trigger a cleanup workflow for derived data, without waiting for any identity-provider migration to complete. A minimal compliant architecture requires four components: an identity adapter, a consent ledger, a policy function, and an audit sink. The design is intended to remain portable across authentication providers, though teams must consult legal counsel to determine jurisdiction-specific consent language and data retention requirements.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How to Keep Insurance Claim Retrieval Accurate Across Document Revisions

Insurance claims systems that ingest updated PDFs without proper versioning risk surfacing outdated figures—such as superseded damage estimates—to adjusters with false confidence. The core fix relies on three principles: assigning every document chunk an immutable identity based on a content digest, keeping only one revision answerable at any time, and applying each revision change as an atomic commit. Three architectural approaches exist for managing live revisions—query-time filtering, delete-and-insert per document, and full index rebuild with alias swapping—each trading retrieval quality against write and latency costs differently. A control table tracking which digest is currently live per claim and document role serves as the authoritative source, while the vector index must constrain retrieval to that live set. Returning an answer that cites a withdrawn figure in an insurance workflow is treated as a compliance defect, not a relevance problem, making this a governance issue as much as a technical one.

0 comments Read more at DEV Community