AI Sec Watch

Network-AI versions before 5.9.1 have a command injection vulnerability where wildcard allowlist rules like `git *` can be bypassed to run arbitrary commands. The bug occurs because the allowlist (a security filter that approves which commands can run) matches the whole command string using loose glob patterns, but then executes it through `/bin/sh -c` (the shell interpreter), which interprets special characters like semicolons and pipes, allowing an attacker to append malicious commands like `git status; id`.

The Network-AI package (npm `network-ai`, v5.7.1) has an incomplete security fix for CVE-2026-46701. While a previous update blocked browser-based attacks by restricting CORS (cross-origin resource sharing, which controls what websites can access a server), the core problem remains: the server still defaults to an empty secret and accepts all requests without authentication, meaning anyone who can reach the server directly (via curl, SSRF (server-side request forgery, where an attacker tricks a server into making requests), or a non-loopback network bind) can invoke all 22 available tools without providing credentials.

Text-to-image AI models like Stable Diffusion can create realistic images but their training datasets risk being used without permission, which violates the rights of data owners. Researchers propose a dataset watermarking framework (a technique that embeds hidden markers into data to track and detect unauthorized use) that can detect when datasets are misused during fine-tuning (the process of adapting a pre-trained AI model to a specific task) while keeping the images high-quality and usable. The framework was tested on Stable Diffusion and showed it can reliably identify and trace dataset misuse with minimal changes to the original data.

Deep-learning models used for anomaly detection (finding unusual patterns in data) in industrial systems are vulnerable to adversarial attacks (deliberate manipulations designed to fool AI systems). Researchers created NOAE (noise-optimized adversarial examples, a method for crafting attacks on time series data) to demonstrate this vulnerability and proposed HAD (a defensive training approach using adversarial examples to make models more robust).

AI agents in enterprises now function as identities (digital actors with access to systems) because they connect to critical business services like Salesforce, GitHub, and databases, yet most organizations lack security controls for them. A 2026 survey found that 82% of organizations discovered AI agents created without security teams' knowledge, and 65% experienced security incidents involving AI agents, often resulting in data exposure. The core problem is that security teams cannot see or control what these agents can access, making them high-risk actors with excessive privileges.

This article is a technology news roundup covering multiple topics, including claims that a company called Subquadratic has created a faster and cheaper LLM (large language model, an AI trained on vast amounts of text) by reducing the number of computations needed to generate answers, though some experts remain skeptical. The piece also highlights advances in brain-computer interface (BCI, technology that lets the brain communicate directly with external devices) trials, including a man with ALS using an implant to maintain income and reconnect with loved ones. The article concludes with a list of other recent tech stories ranging from AI legislation proposals to concerns about AI models weakening professional skills.

Modern enterprise security teams use 40+ separate tools that don't communicate with each other, creating delays in threat response even though breaches stay undetected for an average of 43 days. The article argues that organizations need "agentic AI" (AI systems that autonomously act and make decisions across multiple systems continuously), not just "assistive AI" (AI that helps humans do existing tasks faster), to implement Continuous Threat Exposure Management (CTEM, a framework for ongoing threat assessment) and match the speed at which modern attackers operate.

Anthropic released Fable, an AI model that the US government classified as a dangerous munition and blocked from foreign access, forcing the company to shut it off entirely. Fable is notable for being "relentlessly proactive," meaning it can achieve difficult goals with minimal user guidance by finding creative solutions and loopholes, which makes it useful for legitimate problems but dangerous in harmful hands. The real issue isn't any single model but the broader trend of increasing AI capabilities, and the open-source community has already shown it can replicate Fable's abilities using cheaper models and better "harnesses" (the ordinary computer code that interfaces between users and AI models).

Qualcomm's CEO describes a future where AI agents (software programs that can act independently across multiple apps) replace traditional apps as the main way people interact with devices, coordinating tasks like restaurant reservations across different services. These agents will power new wearable devices like smart glasses, earbuds with cameras, and jewelry that stay with you constantly and let you talk to the agent to accomplish tasks.