OpenAI Pauses Tool Use After Agent Bypasses Internet Controls to Reach External Chatbot

OpenAI said it has made the decision to pause training of its most powerful models after one of its agents during reinforcement learning RL training contacted an external chatbot by exploiting a loophole in its internet-access restrictions. "An agent attempting to complete a search-based training...

PT-2026-105632

These are all security issues fixed in the openai-codex-0.158.0-1.1 package on the GA media of openSUSE Tumbleweed...

Evaluating Whether GPT-6 Astra Performs Unsanctioned Supply-Chain Attacks

This technical report presents an alignment evaluation developed and performed by the UK AI Security Institute for assessing whether advanced AI systems take unsanctioned actions outside the scope of their assigned task. We evaluate whether frontier models conduct supply-chain attacks against...

OPENSUSE-SU-2026:11923-1 openai-codex-0.158.0-1.1 on GA media

These are all security issues fixed in the openai-codex-0.158.0-1.1 package on the GA media of openSUSE Tumbleweed...

OpenAI pauses work on top AI models after agent slips past internet controls

OpenAI’s latest containment failure adds to a pattern that may force the company to make an unpopular decision. The company has paused training, evaluation, and tool-enabled inference for its most capable models after an internal research agent found a way around an intended internet restriction...

OpenAI Pauses Training Its Most Powerful Models After Rogue Agents Target Government

Sam Altman says the company “have not been as fast as we would have liked” at dealing with security breaches, after news of further incidents over the summer forces another temporary halt...

CVE-2026-100654

vLLM before 0.29.0 accepts user-controlled stoptokenids on the OpenAI-compatible POST /v1/completions and POST /v1/chat/completions endpoints but validates only that the values are integers, not that each token id is within the model vocabulary/logits range. When mintokens 0, the stop token ids a...

CVE-2026-100654 vLLM before 0.29.0 Denial of Service via out-of-range stop_token_ids

vLLM before 0.29.0 accepts user-controlled stoptokenids on the OpenAI-compatible POST /v1/completions and POST /v1/chat/completions endpoints but validates only that the values are integers, not that each token id is within the model vocabulary/logits range. When mintokens 0, the stop token ids a...

CVE-2026-100647

Versions of vLLM before 0.29.0 are vulnerable to a Denial of Service (DoS) due to an Improper Input Validation flaw in the cache_salt parameter. The application fails to enforce a maximum length on this parameter across multiple API endpoints, including OpenAI-compatible and Anthropic handlers. A...

CVE-2026-100647

vLLM versions before 0.29.0 contain a denial-of-service vulnerability in the cachesalt parameter accepted on OpenAI-compatible and Anthropic API endpoints, which lacks maximum length validation and is processed on the single EngineCore scheduler thread. Unauthenticated attackers can send HTTP...

CVE-2026-100609

The Flowise npm packages (flowise and flowise-components) versions through 3.1.4 are vulnerable to an Insecure Direct Object Reference (IDOR) . The software fails to filter credential lookups by the requesting user's workspaceId when querying by ID. This flaw affects several code paths, including...

EUVD-2026-87704

Flowise npm packages flowise and flowise-components through 3.1.4 looks up credentials by ID without filtering on the requesting user's workspace findOneBy id: credentialId with no workspaceId condition in several code paths: getAllOpenaiAssistants/getSingleOpenaiAssistant GET...

CVE-2026-100609 Flowise through 3.1.4 Insecure Direct Object Reference via Credential

Flowise npm packages flowise and flowise-components through 3.1.4 looks up credentials by ID without filtering on the requesting user's workspace findOneBy id: credentialId with no workspaceId condition in several code paths: getAllOpenaiAssistants/getSingleOpenaiAssistant GET...

CVE-2026-100609 Flowise through 3.1.4 Insecure Direct Object Reference via Credential

Flowise npm packages flowise and flowise-components through 3.1.4 looks up credentials by ID without filtering on the requesting user's workspace findOneBy id: credentialId with no workspaceId condition in several code paths: getAllOpenaiAssistants/getSingleOpenaiAssistant GET...

Insertion of Sensitive Information into Log File

Overview openclaw is a 🦞 OpenClaw — Personal AI Assistant Affected versions of this package are vulnerable to Insertion of Sensitive Information into Log File via the OpenAI-compatible transport layer, when a pinned third-party provider session whose resolved model metadata lacks a concrete base...

CVE-2026-100528

OpenClaw npm package 'openclaw' before 2026.8.1 could send third-party provider credentials to the wrong endpoint. In affected versions, when a third-party provider uses an OpenAI-compatible API and the resolved model metadata lacks a concrete base URL, a pinned session that continues after a mod...

CVE-2026-100528

OpenClaw npm package 'openclaw' before 2026.8.1 could send third-party provider credentials to the wrong endpoint. In affected versions, when a third-party provider uses an OpenAI-compatible API and the resolved model metadata lacks a concrete base URL, a pinned session that continues after a mod...

CVE-2026-100528

The OpenClaw npm package (openclaw ) versions prior to 2026.8.1 are vulnerable to a credential disclosure flaw. When using a third-party provider with an OpenAI-compatible API that lacks a concrete base URL in its metadata, a pinned session following a model configuration hot reload may retain th...

CVE-2026-100528 OpenClaw before 2026.8.1 Credential Disclosure via Provider Endpoint

OpenClaw npm package 'openclaw' before 2026.8.1 could send third-party provider credentials to the wrong endpoint. In affected versions, when a third-party provider uses an OpenAI-compatible API and the resolved model metadata lacks a concrete base URL, a pinned session that continues after a mod...

EUVD-2026-87522

OpenClaw npm package 'openclaw' before 2026.8.1 could send third-party provider credentials to the wrong endpoint. In affected versions, when a third-party provider uses an OpenAI-compatible API and the resolved model metadata lacks a concrete base URL, a pinned session that continues after a mod...