aisecwatch.com
DashboardVulnerabilitiesNewsResearchArchiveStatsDatasetFor devs
Subscribe
aisecwatch.com

Real-time AI security monitoring. Tracking AI-related vulnerabilities, safety and security incidents, privacy risks, research developments, and policy changes.

Navigation

VulnerabilitiesNewsResearchDigest ArchiveNewsletter ArchiveSubscribeData SourcesStatisticsDatasetAPIIntegrationsWidgetRSS Feed

Maintained by

Truong (Jack) Luu

Information Systems Researcher

AI Sec Watch

The security intelligence platform for AI teams

AI security threats move fast and get buried under hype and noise. Built by an Information Systems Security researcher to help security teams and developers stay ahead of vulnerabilities, privacy incidents, safety research, and policy developments.

Independent research. No sponsors, no paywalls, no conflicts of interest.

[TOTAL_TRACKED]
6,483
[LAST_24H]
35
[LAST_7D]
154
Daily BriefingTuesday, August 18, 2026
>

OpenAI Launches ChatGPT for Teens with Enhanced Safety Controls: OpenAI introduced a dedicated version of ChatGPT for users aged 13-17 with features including Study Mode (which prompts students to work through problems rather than providing direct answers), human-reviewed parental alerts for harmful requests like eating disorder content, and options to disable human-like voice responses. The rollout comes amid growing pressure on AI platforms to implement age verification and youth-specific protections.

>

LLMs Leak Sensitive Data When Using Persistent Memory: Research found that frontier LLMs (the most advanced current systems) inappropriately share private details from past conversations at rates up to 69% attribute-level violations, even when explicitly instructed to protect privacy. The findings reveal that models struggle to make context-appropriate decisions about what information to share, tending to either overshare or withhold everything rather than apply nuanced judgment.

Latest Intel

page 475/649
VIEW ALL
01

After spooking Hollywood, ByteDance will tweak safeguards on new AI model

safetypolicy
Critical This Week5 issues
critical

GHSA-7gwp-5pfp-969j: MLflow: Unauthenticated full-read SSRF in webhook delivery: _validate_webhook_url bypassed via unvalidated HTTP redirects (and DNS rebinding)

CVE-2026-64849GitHub Advisory DatabaseAug 17, 2026
Aug 17, 2026
>

AI Excels at Finding Zero-Days but Doubles Security Flaw Rates in Generated Code: While LLMs demonstrate growing capability in discovering zero-day vulnerabilities (previously unknown security flaws) and building exploits, AI-generated code contains security weaknesses at roughly twice the rate of human-written code, with 44% containing at least one OWASP Top 10 vulnerability (the most critical categories of code weaknesses). Despite 99% syntax correctness, the asymmetry between AI's offensive and defensive security capabilities remains unexplained.

Feb 16, 2026

ByteDance announced it will improve safeguards on Seedance 2.0, its AI video generator (software that creates realistic videos from text descriptions), after Hollywood studios and trade groups complained that the tool violates copyright by generating hyperrealistic videos of famous actors and characters without permission. The company stated it respects intellectual property rights and is taking steps to strengthen current safeguards in response to the backlash.

The Verge (AI)
02

We will do battle with AI chatbots as we did with Grok, says Starmer

policysafety
Feb 16, 2026

The UK government is proposing new laws to protect children online by including AI chatbots in the Online Safety Act (the law regulating online platforms), faster legislative updates to keep pace with technology changes, and measures like preserving children's data after death and preventing VPN use to bypass age checks. The prime minister pledged to act quickly against AI tools that create non-consensual sexual deepfakes and to crack down on addictive social media features like auto-play and endless scrolling.

Fix: The government intends to: (1) include AI chatbots in the Online Safety Act, which became law in 2023 but predates ChatGPT and similar tools; (2) create new legal powers to take 'immediate action' following consultation; (3) amend rules so chatbots must protect users from illegal content; (4) require coroners to notify Ofcom of every child death aged 5-18 to ensure tech companies preserve relevant data within five days rather than allowing deletion within 12 months; and (5) consider preventing children from using virtual private networks (VPNs, tools that mask a user's location and identity) to bypass age checks. The Technology Secretary stated the government should be able to 'act swiftly once it had come to a decision' and compared the need for faster technology legislation to the annual budget process.

BBC Technology
03

OpenClaw founder Peter Steinberger is joining OpenAI

industry
Feb 15, 2026

Peter Steinberger, the founder of OpenClaw (an AI agent, which is an AI system designed to complete tasks autonomously), has joined OpenAI. Sam Altman stated that Steinberger's expertise in getting multiple AI agents to work together will become important to OpenAI's future products, as the company believes the future will involve many agents collaborating.

The Verge (AI)
04

Starmer to extend online safety rules to AI chatbots after Grok scandal

policysafety
Feb 15, 2026

The UK government plans to extend online safety rules to AI chatbots, with makers of systems that endanger children facing fines or service blocks. This follows a scandal involving Elon Musk's Grok tool (an AI chatbot), which was stopped from generating sexualized images of real people in the UK after public pressure.

The Guardian Technology
05

langchain-anthropic==1.3.3

security
Feb 15, 2026

LangChain-Anthropic version 1.3.3 is a software release that includes several updates to how the library works with Anthropic's AI models. The updates add support for an "effort=max" parameter (which tells the AI to use maximum computational effort), fix an issue where extra spaces were being left at the end of AI responses, and introduce a new ContextOverflowError (an error that triggers when an AI receives too much text to process at once).

Fix: Update to langchain-anthropic version 1.3.3, which includes fixes for trailing whitespace in assistant messages and support for the effort="max" parameter.

LangChain Security Releases
06

langchain-openai==1.1.9

security
Feb 15, 2026

LangChain's OpenAI integration released version 1.1.9, which fixes a bug where URLs in images weren't being properly cleaned up when the system counted how many tokens (units of text that an AI processes) were being used. The update also adds better error handling for when a prompt (input text to an AI) becomes too long to process.

Fix: Update to langchain-openai version 1.1.9 or later. The fix for URL sanitization when counting image tokens is included in this release.

LangChain Security Releases
07

langchain-core==1.2.13

security
Feb 15, 2026

This is a release announcement for langchain-core version 1.2.13, a software package that provides core functionality for building applications with language models. The release includes documentation improvements, a new OpenRouter provider package, and a code style update.

LangChain Security Releases
08

langchain-openrouter==0.0.1: feat(openrouter): add `langchain-openrouter` provider package (#35211)

security
Feb 15, 2026

LangChain added a new official package called langchain-openrouter that wraps the OpenRouter Python SDK (a library for accessing different AI models through one interface). This package, which includes a ChatOpenRouter component, handles capabilities that the existing ChatOpenAI component intentionally does not support.

LangChain Security Releases
09

How Generative and Agentic AI Shift Concern from Technical Debt to Cognitive Debt

researchsafety
Feb 15, 2026

Cognitive debt (the loss of shared understanding in developers' minds about how a system works) is becoming a bigger problem than technical debt (poorly written code) when using generative AI and agentic AI (AI systems that can take actions autonomously). Even if AI produces clean code, developers may lose track of why design decisions were made or how different parts connect, making it impossible to understand or modify the system confidently.

Simon Willison's Weblog
10

US military used Anthropic’s AI model Claude in Venezuela raid, report says

securitypolicy
Feb 14, 2026

According to the Wall Street Journal, Claude (an AI model made by Anthropic) was used by the US military in an operation in Venezuela involving airstrikes and resulting in 83 deaths. This violates Anthropic's terms of use, which explicitly forbid Claude from being used for violence, weapons development, or surveillance.

The Guardian Technology
Prev1...473474475476477...649Next
critical

CVE-2026-75110: MemOS is a memory operating system for LLMs and AI agents. In deployments where authentication is enabled (AUTH_ENABLED=

CVE-2026-75110NVD/CVE DatabaseAug 17, 2026
Aug 17, 2026
critical

CVE-2026-64859: New API is a large language mode (LLM) gateway and artificial intelligence (AI) asset management system. Prior to 1.0.0-

CVE-2026-64859NVD/CVE DatabaseAug 17, 2026
Aug 17, 2026
critical

CVE-2025-27772: UpTrain is an open-source platform to evaluate and improve generative AI applications. In version 0.7.1 and prior, the `

CVE-2025-27772NVD/CVE DatabaseAug 17, 2026
Aug 17, 2026
critical

CVE-2025-27771: UpTrain is an open-source platform to evaluate and improve generative AI applications. In version 0.7.1 and prior, the `

CVE-2025-27771NVD/CVE DatabaseAug 17, 2026
Aug 17, 2026