Pydantic AI is a Python agent framework for building applications and workflows with Generative AI. From 1.77.0 until 1.107.6 and 2.44.0, the local web_fetch_tool and the WebFetch local fallback compare blocked_domains entries with a URL hostname before both values are normalized to the form used by getaddrinfo. An attacker-influenced model can use an equivalent IDNA spelling, non-ASCII label separator, case variation, or trailing root label that resolves to a blocked host but does not match the configured string, causing the application to fetch that host with its own privileges. allowed_domains fails closed for unmatched spellings, and private-IP and cloud-metadata protections remain effective. This issue is fixed in versions 1.107.6 and 2.44.0.
Dislocker versions up to 0.7.3 contain an integer underflow vulnerability in the get_vmk() and get_fvek() functions. This flaw allows attackers to cause out-of-bounds heap reads by supplying a malicious BitLocker volume image with a datum_size smaller than the expected 36-byte AES-CCM header. The vulnerability can lead to application crashes due to over-read in the hexdump() function.
Dislocker versions up to 0.7.3 contain a heap out-of-bounds read vulnerability in the functions get_dataset() and get_next_datum(). This flaw arises because the software does not validate dataset and datum sizes against metadata allocation, allowing crafted BitLocker volume images with inflated sizes to cause crashes or disclose adjacent heap memory.
CVE-2026-105830 is a high-severity denial of service vulnerability in thephpleague commonmark library versions 2.0.0 up to but not including 2.10.2. The flaw exists in the GitHub Flavored Markdown Table extension's TableStartParser::tryStart() method, where an unauthenticated attacker can cause excessive CPU consumption by submitting specially crafted input. This triggers repeated full-buffer scans, leading to quadratic-time resource exhaustion in PHP workers.
League CommonMark from 1.3.0 before 2.10.2 contains a cross-site scripting vulnerability that allows users posting Markdown to bypass the DisallowedRawHtml extension by ending raw HTML with a bare disallowed tag name. Attackers can place a lone <script or <iframe line followed by a block supplying attributes like src or onload, executing stored scripts in viewers' browsers under default GFM settings.
CVE-2026-107287 is a medium severity vulnerability in pydantic-ai affecting versions from 1.77.0 up to but not including 1.107.7, and from 2.0.0b1 up to but not including 2.52.0. The issue involves uncontrolled resource consumption during HTML-to-Markdown conversion of attacker-controlled HTML with deeply nested block elements, causing excessive CPU and memory use. This can delay processing of other tasks. The vulnerability is fixed in versions 1.107.7 and 2.52.0.
> This issue was posted by Codex Desktop using gpt-6.1-sol on behalf of David.
### Summary
Applications that wrap a model with `ConcurrencyLimitedModel` or `limit_model_concurrency` can permanently lose shared concurrency capacity when a streamed request releases its slot from a different task than the one that acquired it. This can happen when a stream ends early, and also when a stream is fully consumed using the default `stream_text()` debouncing.
In an application that exposes an affected streaming endpoint to network clients and shares a long-lived model limiter across requests, a client can repeatedly start a stream and disconnect. The completed requests retain their slots, eventually preventing subsequent requests that share the limiter from proceeding.
Agent-level `max_concurrency` and non-streaming model requests are not affected by this defect.
### Details
The built-in limiter uses `anyio.CapacityLimiter`, which associates each acquired slot with its borrowing task. Pydantic AI's streaming lifecycle can acquire the slot on the task consuming the stream and run cleanup on another internal task. The limiter rejects that release, so the slot remains occupied even after the request has ended. Cleanup can raise a `RuntimeError`; a later request on the borrowing task can also fail because that task still holds a slot.
Early termination includes stopping iteration, a consumer exception, and cancellation. Fully consuming `stream_text()` with its default `debounce_by=0.1` can also reach the cross-task release path. Fully consumed streams must therefore not be assumed safe.
### Mitigation
Upgrade to a patched release of `pydantic-ai` or `pydantic-ai-slim`. If you cannot upgrade yet, use the agent-level `max_concurrency` setting instead of a concurrency-limited model, or avoid streaming runs through a concurrency-limited model.
CVE-2026-107586 is a medium severity vulnerability in Progressive Robot Ltd hMailServer versions 6.2.28 through 6.3.5. It involves uncontrolled eviction in the browser session table of the REST API, allowing a remote authenticated user to forcibly end other users' sessions by filling the session table. This occurs because the session table, shared across all accounts and administrative sessions, evicts the least recently used session when full without limiting the number of sessions per account. The vulnerability can cause denial of service for webmail and administration sessions by repeatedly signing in and filling the session table.
CVE-2026-107585 is a medium severity vulnerability in Progressive Robot Ltd's hMailServer versions 6.3.4 and 6.3.5. It involves uncontrolled eviction in the pending sign-in tables of the REST API, allowing a remote unauthenticated attacker to disrupt other users' OpenID Connect, SAML, and passkey sign-ins. Attackers can repeatedly initiate sign-ins to evict legitimate pending sign-ins, causing denial of single sign-on and passkey authentication for affected users.
mJobTime through build 15.7.3.32 contains an unauthenticated SQL execution vulnerability in the Login.aspx admin panel handlers, where the runQueryButton postback and exportSqlQuery_Server PageMethod execute caller-supplied SQL against the backing Sybase SQL Anywhere database using DBA/sysadmin privileges with no server-side authentication enforced beyond a client-side sessionStorage flag. Attackers can submit arbitrary SQL through these exposed endpoints to invoke xp_cmdshell and xp_read_file, achieving pre-authentication remote code execution as LocalSystem via a single HTTP request.
This content discusses which certifications currently satisfy the Department of Defense (DoD) CSSP Analyst requirement under DoD 8140/8570. It highlights conflicting information from various online sources regarding acceptable certifications such as CEH, CySA+, GCIA, GCIH, SCYBER, and CFR. The author seeks clarification from individuals familiar with the DoD requirements and notes difficulty accessing official DoD information due to CAC access issues.
A flaw was found in the OIDC Dynamic Client Registration (DCR) component of Keycloak. A bug in the response serialization causes the backchannel logout offline token revocation setting to be omitted from responses. When a client performs a standard update, this missing information causes the setting to be silently disabled. As a result, offline tokens may remain valid even after a user session is terminated via backchannel logout.
A flaw was found in the installation provider and client registration endpoints of the Keycloak identity management service. A realm administrator with only the read-only view-clients role can access the active primary secret of any confidential client, which should normally be restricted. This exposed secret can be used to impersonate the client and gain unauthorized access to its associated service account permissions.
A flaw was found in luksmeta. A local attacker with administrative privileges can cause data corruption when saving metadata to a Linux Unified Key Setup (LUKS) device. Due to incorrect boundary calculations and flawed overlap detection, new metadata entries can be written beyond available free space or over existing records. This issue can corrupt stored encrypted payload data or existing metadata, potentially rendering the affected data inaccessible.
CVE-2026-12859 is a missing authorization vulnerability in Caz Informatics Services Trade Inc.'s Advancity ALMS Cloud product. It allows unauthorized access to functionality that is not properly constrained by access control lists (ACLs). The vulnerability affects versions up to and including 2026-10-08. The vendor was contacted but did not respond to the disclosure. The CVSS 3.1 base score is 6.5, indicating a medium severity issue with network attack vector, low attack complexity, and requiring low privileges without user interaction. Confidentiality impact is high, but integrity and availability are not affected.
CVE-2026-93017 is a privilege management vulnerability in Red Hat OpenShift Container Platform 4.14. The insights-operator-gather ClusterRole grants its service account unrestricted read access to all secrets across all namespaces. An attacker with the ability to spawn a pod using this service account can access any secret in the cluster.
IBM DataPower Gateway versions 10.5.0.0 through 10.5.0.22, 10.6.0.0 through 10.6.0.10, 10.6.1 through 10.6.6, and 11.0.0.0 through 11.0.0.2 contain a heap-based buffer overflow vulnerability due to improper bounds checking. This flaw allows an unauthenticated remote attacker to overflow the buffer and execute arbitrary code on the affected system. The vulnerability has a critical severity with a CVSS score of 9.8, indicating high impact on confidentiality, integrity, and availability.
This content discusses the current state and effectiveness of security awareness training in enterprises. It highlights that while all organizations conduct such training, its tangible benefits are debated due to repetitive, generic content and compliance-driven approaches. The rise of AI-enabled social engineering attacks challenges traditional training methods. Experts suggest that awareness training needs to be more frequent, context-driven, behavioral-focused, and integrated with technical controls rather than relied upon as the sole defense. The article emphasizes that training alone cannot counter modern threats, especially those leveraging AI for sophisticated phishing and social engineering.
CVE-2026-105331 is a local privilege escalation vulnerability in Checkmk versions 2.5.0 up to but not including 2.5.0p10. It arises from an untrusted search path issue in the 'mk-oracle' agent plugin that references the Oracle Instant Client. A user with access to modify the Oracle Instant Client used by this plugin can escalate privileges if the plugin is enabled on the agent.
Integrics Enswitch 3.13 through 4.4 contains an authentication bypass vulnerability in /api/json/user/password/update/ that allows unauthenticated attackers to change account passwords by omitting the reset parameter. Attackers can target accounts with no pending reset, whose empty reset_key matches the defaulted empty value, to take over administrator accounts after enumerating valid usernames.