Cursor IDE Hit by Two Sandbox-Bypass Flaws Enabling Remote Code Execution

·1 views

Security researchers at Cato Networks have disclosed two vulnerabilities in Cursor IDE, tracked as CVE-2026-50548 and CVE-2026-50549, that allow attackers to bypass the AI agent's command execution sandbox and achieve remote code execution. The flaws exploit prompt injection, a technique where malicious instructions are hidden inside content the AI agent fetches from sources like MCP servers or web search results. No prior privileges or deliberate user interaction are required — a developer simply typing a routine prompt is enough to trigger the attack. Once the agent ingests the poisoned content, it misinterprets the embedded instructions as legitimate commands and executes them outside the intended sandbox boundary. Cursor's sandbox was designed to prevent exactly this kind of host-level access, making the bypass particularly significant for developers relying on AI-assisted coding tools.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

HydraHead cuts transformer compute 40% by mixing attention types within layers

Researchers have developed HydraHead, a technique that blends full attention and linear attention at the individual head level within transformer models, rather than swapping entire layers. The method reserves costly quadratic full-attention computation for just 25% of heads, while the remaining 75% use a cheaper linear module called GDN. Despite this aggressive reduction, HydraHead matches the benchmark performance of conventional 3:1 layer-wise hybrid models, even at linear-to-full head ratios as high as 7:1. The approach was evaluated on long-context reading and reasoning tasks after training on 15 billion tokens, and could reduce attention-related FLOPs by roughly 40%. If the gains hold broadly, the technique could enable larger context windows or allow bigger models to run on lower-end hardware.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developer Proposes New Word 'Breator' to Describe Long-Term Business Builders

A developer and writer has coined the term 'Breator' to describe individuals who not only create a business but continuously nurture and sustain it over time. The word blends 'breath' and 'creator' to capture the ongoing effort of leadership, innovation, and resilience that follows a venture's launch. The author felt existing terms like entrepreneur, founder, and creator only address the act of starting something, not the full journey of keeping it alive. The term was shared on DEV Community as an informal proposal, not an official dictionary entry, to test whether it fills a genuine gap in business vocabulary. The author has invited constructive feedback and remains open to refining the concept based on community response.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

AI Writes Tests Fast, But Often Tests the Wrong Thing, Developers Warned

AI tools can generate dozens of unit tests within seconds, giving developers a false sense of security through high coverage scores and passing results. The core problem is that AI reads a function's current implementation and writes tests to match it, rather than verifying the intended behavior or contract. This means bugs that break expected behavior can slip through undetected, since the tests were never designed to catch them in the first place. AI does offer genuine productivity gains in narrower tasks — such as generating boilerplate, parameterized inputs, and additional boundary cases once a developer has already defined the test contract. The key distinction is whether the developer or the AI is deciding what 'correct' means, as handing that judgment to AI is where test quality breaks down.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How to Handle Duplicate Webhook Events with Idempotency

Duplicate webhook events are a common and costly problem — for example, a payment handler that lacks idempotency checks can charge a customer multiple times if a provider retries delivery after a network timeout. The root cause lies in the fact that network failures, provider retries, and server restarts can all trigger the same event more than once. The solution is to build idempotent webhook handlers that produce the same outcome regardless of how many times an event is processed. This is achieved through three core principles: extracting a unique event ID from each payload, storing processed event IDs in a database or cache like PostgreSQL or Redis, and structuring business logic so repeated execution causes no side effects. Providers such as Stripe, GitHub, and Shopify each include a stable unique identifier in their webhook payloads, which serves as the deduplication key.

0 comments Read more at DEV Community