xAI's Grok Bot Beta Shares One Cloud Computer Across All Bots, Raising Security Questions

·3 views

xAI launched Grok Bot in beta on 11 August 2026 as a separate product from its Grok chat assistant. These persistent AI agents run on a cloud machine with a browser, filesystem and terminal, and can sign into existing tools to complete multi-step tasks. Although xAI's launch material says each Bot has its own computer, its documentation states that all Bots on an account share one persistent machine and should not be treated as a security boundary. Because sessions and logins are shared across Bots, reviewers note that access must be scoped to what every Bot on the account may touch. The analysis also points out that Grok Bot is cloud-hosted with no on-premises deployment, making it better suited to low-sensitivity workloads than to tasks needing strict perimeter control.

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)

Why Input Lag Can Mislead Developers: The Real Causes Behind Sluggish UIs

A developer describes users reporting sluggish input in a web app, where profiling pointed at JavaScript but the actual bottleneck was more complicated. The article explains that browsers pass input events through several stages, including OS processing, event coalescing, dispatch, and handler execution, any of which can introduce delay. It then outlines the rendering pipeline that follows JavaScript work, covering style recalculation, layout, paint, and compositing, which can postpone visible updates. The author notes that common DevTools signals such as Event Timing, Frames, and Long Tasks are easy to misread, since a handler may run late because of earlier work. Input latency therefore includes both event delay and render delay, and diagnosing it requires looking at the full pipeline rather than only the longest JavaScript task.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Meta open-sources code for building Muse AI gadgets

Meta has released open-source code that lets developers build gadgets powered by its Muse AI model. The article describes a builder who used the Amblyotube repository and the Muse model on a Raspberry Pi to create a gaze-tracking therapy setup for amblyopia. Muse's compact, quantized design allowed on-device inference at around 60 frames per second with a model under 4 MB. Fine-tuning on a few labeled sessions reportedly reduced calibration error from 2.4 degrees to 0.7 degrees. The author encourages others to assemble similar projects from the publicly available code on GitHub.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Gating agent-written patches on fixture sensitivity instead of green test suites

A passing test suite can be weak evidence for an agent-authored patch, since tests may pass because they have stopped constraining behavior. The article proposes measuring how many deliberate breakages a fixture still catches, rather than only whether tests pass. Its approach combines property obligations, canary mutations, and a freeze registry with expiry, wired into a pre-merge check. The main risk described is fixture drift, where fixtures and assertions gradually loosen as patches are added. The gate is designed to detect this loss of test strength without assuming malicious intent.

0 comments Read more at DEV Community

ProgrammingHacker News ·

Essay Argues Writing Is Closer to Thought Than Code

A blog post by Daniel Miessler contends that writing is more closely related to the process of thinking than to the act of coding. The piece appeared on Hacker News, where it received 3 points and no comments at the time of capture. The source material provided is limited to the headline, URL, and engagement metrics, so the full argument is not summarized here. Readers would need to consult the original article for the author's specific reasoning and examples.

0 comments Read more at Hacker News