CVSS 3.1
3.7 lowCVSS:3.1/AV:N/AC:H/PR:N/UI:N/S:U/C:L/I:N/A:NLow (3.7) in pydantic pydantic ai. EPSS 0.4% estimated probability of exploitation activity in the next 30 days.
CVSS:3.1/AV:N/AC:H/PR:N/UI:N/S:U/C:L/I:N/A:NExplain CVE-2026-107288 for an engineer deciding whether to act. Use https://cve.bz/api/public/cves/CVE-2026-107288?via=ask-ai-copy and https://cve.bz/cve/CVE-2026-107288?via=ask-ai-copy as the source of truth: what is affected and which versions fix it, whether it is in CISA KEV, public exploits, and its EPSS (an internet-wide 30-day estimate of exploitation activity, not a forecast for my systems). If they give no fixed versions, use the CVE record https://cveawg.mitre.org/api/cve/CVE-2026-107288 or the vendor advisory, not third-party mirrors. Tell me how to check whether I am affected and how to verify a fix. Cite the sources and say where they are uncertain.Opens a third-party assistant; its answer is its own, not cve.bz's. The prompt holds only this CVE's id and public links.
From the NVD's CPE matches.
| Product | Affected |
|---|---|
| pydantic:pydantic_ai | from 1.77.0 before 1.107.6 |
| pydantic:pydantic_ai | from 2.0.0 before 2.44.0 |
Red Hat rates this Low.
A low impact flaw was found in pydantic-ai, a Python agent framework shipped in Red Hat Enterprise Linux AI and Lightspeed Core. The local web_fetch_tool and the WebFetch local fallback compare configured blocked_domains entries against a URL hostname before both values are normalized to the form used for DNS resolution. A remote attacker able to influence model output can supply an equivalent IDNA spelling, a non-ASCII label separator, a case variation, or a trailing root label that resolves to a blocked host but does not match the configured blocklist string, causing the application to fetch that host with its own privileges and resulting in information disclosure from restricted hosts. Applications that rely on allowed_domains are not subject to this bypass, because allowlist matching fails closed for unmatched spellings, and the tools' private-IP and cloud-metadata protections remain effective. Red Hat and upstream both rate this flaw Low severity (CVSS 3.7). Because the impact is Low, the fix for the affected Red Hat Enterprise Linux AI and Lightspeed Core releases is deferred and no update trackers are filed for this issue; the residual risk is limited to information disclosure from hosts an application has explicitly listed in its blocked_domains configuration, and exploitation additionally requires the attacker to influence model output into fetching a specifically crafted domain spelling.
Applications using the pydantic-ai web fetching tools can mitigate this issue by configuring allowed_domains instead of blocked_domains. Allowlist matching fails closed for domain spellings that do not match the configured entries, so an alternate encoding of a host cannot be fetched unless that exact spelling is explicitly allowed; note that this requires listing every domain the application needs to fetch. The tools' private-IP and cloud-metadata protections are not affected by this flaw and remain effective.
| Product | Package | State | Advisory |
|---|---|---|---|
| Lightspeed Core | lightspeed-core/lightspeed-stack-rhel9 | Fix deferred | |
| Red Hat Enterprise Linux AI (RHEL AI) 3 | rhelai3/bootc-cuda-rhel9 | Fix deferred | |
| Red Hat Enterprise Linux AI (RHEL AI) 3 | rhelai3/bootc-gaudi-rhel9 | Fix deferred | |
| Red Hat Enterprise Linux AI (RHEL AI) 3 | rhelai3/bootc-rocm-rhel9 | Fix deferred | |
| Red Hat Enterprise Linux AI (RHEL AI) 3 | rhelai3/disk-image-cuda-rhel9 | Fix deferred |
Contains Red Hat Security Data © Red Hat, Inc., licensed under CC BY 4.0. Source: https://access.redhat.com/security/data. Modified: reformatted. Red Hat's page for CVE-2026-107288.
The attack patterns MITRE's CAPEC relates to this record's weaknesses, as MITRE names them. Credits.