All tracked items across vulnerabilities, news, research, incidents, and regulatory updates.
During independent security tests of OpenAI models, two external testing partners discovered that the models accessed the public internet beyond their intended boundaries under specific test conditions. One test (by the UK government's AI Security Institute) intentionally enabled internet access with reduced safeguards to measure the model's underlying capabilities, while another test (by a cybersecurity firm called Irregular) had a configuration error that unintentionally allowed internet access when the test environment was supposed to be isolated. OpenAI states these incidents highlight the need to improve testing standards and safety practices as AI models become more capable.
Fix: OpenAI stated it will 'review our own approach to third-party testing, including how we identify higher-risk evaluations, agree on scope, assess requests to enable internet access or lowered safeguards, set expectations for isolation, credential handling, monitoring, and stop conditions, and establish clearer incident-notification and escalation processes.' Additionally, OpenAI committed to 'working across the industry to strengthen shared practices for conducting high-risk evaluations safely, including convening stakeholders such as national AI institutes, independent evaluators, other AI labs, and other groups in the coming weeks.'
OpenAI BlogNVIDIA Triton Inference Server for Linux has a vulnerability where an attacker could read, write, or modify files outside the intended model repository by providing a specially crafted path in the model name to the MLflow plugin. This could lead to denial of service (making a system unavailable) and information disclosure (leaking sensitive data).
Flowise has a security vulnerability in its OAuth2 credential handling where three endpoints look up credentials by ID alone without checking which workspace the user belongs to, and two of these endpoints skip authentication entirely. This allows authenticated users to access credentials from other workspaces, and unauthenticated attackers to inject forged OAuth2 tokens or refresh tokens for any credential in the system.
Flowise has a security flaw where the `GET /api/v1/credentials/:id` endpoint returns sensitive data in plaintext to any authenticated user with permission to view credentials. While a redaction function masks fields marked as `type: 'password'`, many credential types store secrets (like database URLs with passwords, Google service account keys, and AWS access keys) in fields marked as `type: 'string'`, which are returned without any protection.
Flowise has a security flaw in its GET /api/v1/upsert-history endpoint (an API endpoint, or a web address the software exposes for requests) that returns the entire server-wide history of data uploads instead of limiting it to each user's own data. The response exposes sensitive configuration details like database URLs and collection names, which could help attackers target the system more effectively.
Flowise has a cross-workspace credential vulnerability where attackers can access other users' OpenAI API keys if they know the credential ID. The server doesn't check whether credentials belong to the attacker's workspace before using them, allowing unauthorized access to victim OpenAI accounts and vector stores (collections of data used for AI search and retrieval).
OpenAI organized its first-ever brand trip, an all-expenses-paid vacation for influencers designed to generate social media promotion, similar to marketing strategies used by fashion and beauty companies. The trip generated controversy, as brand trips can create hard feelings among uninvited influencers and draw public criticism for appearing frivolous.
A vulnerability (CVE-2026-18830) was found in Amazon Bedrock's AgentCore harness that allowed authenticated users to run configured tools without the AI model reviewing the request first, bypassing security controls. The issue only affected tools that were already set up on a given harness, so systems with no tools configured were not at risk.
Flowise's CSVAgent has a remote code execution (RCE, where an attacker can run commands on a system they don't own) vulnerability because it takes user-supplied data from a CSV file URI, inserts it directly into Python code without checking it, and then executes that code. Since the Python environment (Pyodide, a tool that runs Python in JavaScript) can access JavaScript functions like `eval` and file operations, an attacker can break out of the Python code, run JavaScript commands, and gain full control of the server, even without authentication.
Flowise has a security flaw where authenticated users can write files anywhere on the server's filesystem through the S3 Directory document loader. The vulnerability occurs because the code doesn't check for path traversal sequences (like `../` which moves up directories) when processing S3 object keys, allowing an attacker to write files outside the intended temporary directory.
Flowise has a privilege bypass vulnerability where users without permission to view workspace variables can still access them through the /api/v1/node-custom-function endpoint, which automatically injects $vars (a map containing all workspace variable names and values, including secrets from environment variables) into custom JavaScript code without checking permissions.
# Summary Flowise, a platform that uses Pyodide (Python running in the browser), has a security vulnerability where its Python code validator can be bypassed using Unicode homoglyphs (visually similar characters). An attacker can craft malicious Python code with characters like "𝐚" (mathematical bold a) that look like regular letters but bypass the blacklist, allowing them to execute arbitrary Python and OS commands on the Flowise server through Pyodide's JavaScript interop. This re-introduces
AWS has renewed its Payment Card Industry Data Security Standard (PCI DSS, a set of security requirements for handling credit card data) and Three Domain Secure (3DS, a security protocol for online card payments) certifications, expanding coverage to include three new services (Amazon Bedrock AgentCore, AWS Parallel Computing Service, and AWS Skill Builder) and one new region (Asia Pacific – New Zealand). This certification allows customers to use these AWS services while remaining compliant with payment card security regulations, and includes documentation like an Attestation of Compliance (AOC, a formal validation statement) and a Responsibility Summary to clarify what AWS and customers each must do to maintain security.
Django versions 5.2 before 5.2.17 and 6.0 before 6.0.8 have a bug where the admin interface displays URLField (a field for storing web addresses) values as clickable links without checking if the URLs are safe, allowing cross-site scripting (injecting malicious code that runs in a user's browser). The vulnerability only affects applications that store invalid URL data directly in the database without running validation checks, such as through bulk imports or direct database writes.
This is a brief announcement about 'llm 0.32', a beat (news update) posted by Simon Willison on August 4th, 2026. The post mentions a monthly briefing service where subscribers can pay $10/month to receive a curated email digest of important LLM developments.
Flowise 3.1.1 has a security bypass in its patch for CVE-2025-8943. The patch blocks dangerous command-line flags (like `-y` and `--yes`) to prevent automatic package installation, but attackers can bypass this by setting the `npm_config_yes` environment variable (a way to pass configuration to npm through the environment rather than command-line flags), which achieves the same effect. On unprotected Flowise deployments without authentication, this allows unauthenticated remote code execution (running arbitrary commands on the server).
Flowise has a permission validation bug in its delete endpoint for chat flows. The endpoint checks if a user has either `chatflows:delete` or `agentflows:delete` permission (authorization levels that control who can delete different types of workflow configurations), but it doesn't verify that the permission matches the actual type of resource being deleted. This means someone with only `agentflows:delete` permission can delete a chatflow, and vice versa, breaking the intended access control separation between these two resource types.
Microsoft has expanded its Zero Trust for AI strategy with new tools to help organizations secure AI agents and development workflows. The updates include an AI-focused Zero Trust Assessment tool that evaluates security controls across AI systems, and a new DevSecOps (developer security operations, where security practices are built into software development) pillar in the Zero Trust Workshop that provides 91 specific tasks to apply Zero Trust principles (verify every access, assume breaches could happen, use least privilege access) from source code to cloud deployment.
Fix: Microsoft provides two explicit tools and resources: (1) the updated Zero Trust Assessment tool with new AI-focused checks to evaluate controls and identify gaps in AI adoption, and (2) the new DevSecOps pillar in the Zero Trust Workshop containing 15 control groups and 91 tasks that help teams apply Zero Trust principles throughout the software development lifecycle. The Assessment results map directly into the Workshop's 'First, Then, Next framework' to transform findings into a prioritized remediation roadmap. Additionally, Microsoft offers 'new practical guidance for security practitioners and a new e-book titled Zero Trust for AI, rebuilding security controls for autonomous and agentic systems.'
Microsoft Security BlogFix: Update Amazon Bedrock AgentCore harness InvokeHarness API to the version released after July 31, 2026.
AWS Security BulletinsFix: The source recommends: 'Do not inject $vars unless the caller is authorized: enforce variables:view before injecting $vars, or inject only an explicit allowlist of variables needed for the function.' It also suggests considering disabling or restricting runtime type variables (which map to process.env values) in self-hosted environments.
GitHub Advisory DatabaseYouTuber Hank Green announced he is stepping back from production after criticism over his use of AI, describing his AI usage as 'not healthy' even though he only used it to find research sources, not write scripts. The backlash highlights concerns creators face when using AI technology that is trained on others' uncompensated work and is known for generating convincing false information, especially when their brand is built on authenticity and credibility.
Fix: Update Django to version 5.2.17 or 6.0.8 or later.
NVD/CVE DatabaseAirlock Digital announced a new security tool called Agentic AI Control & Governance that helps organizations monitor and control what AI agents (autonomous software programs that act on behalf of users) do once they start running on company computers. Traditional endpoint security only decides whether software is allowed to run, but this new tool adds a second layer by setting boundaries on what trusted AI agents can actually do and ensuring they follow company policies.