Editor preview panes in AI coding tools were found trusted like browsers, leaving some flaws unpatched

·2 views

Security researchers at Forever Security reported on October 29, 2026 that embedded preview panes in AI coding tools such as Cursor, Copilot, Codex, VS Code, and Claude Desktop were trusted as if they were standalone browsers. Five issues were documented, including a code execution path in Cursor and file reads beyond the project folder in several tools. Copilot fixed its address-bar spoofing flaw in version 1.1.29 before the report, while Codex and VS Code were still unpatched at publication. Microsoft and Anthropic declined the reports citing Workspace Trust, and OpenAI told The Register it is working on a fix. Claude Desktop's patch status remains unclear because no fix version was published.

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)

Generating Localised Price Tables From a Single JSON Source

Pricing pages translated into several languages often end up with inconsistent price tables when updates are applied to some versions but not others. The recommended fix is to store prices as structured data and render every language version from that single source. Currency amounts are set explicitly per currency instead of being converted with live exchange rates, which keeps tables stable and intentional-looking. Intl.NumberFormat is used to handle locale-specific symbol placement, separators and spacing. A test loops over every locale and plan to fail the build when a price or translation key is missing, preventing half-updated tables from shipping.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

IBM Ties AI Agents to Named Owners and Expiring Credentials, Kill Switch Still Preview

IBM announced Agent Identity, now generally available, which links each AI agent to an owner, a permission status, and the systems it may access. Its Vault 2.2 product, also generally available, renews agent credentials on a schedule rather than leaving them static. IBM cited Gartner's projection that a typical Fortune 500 company will run more than 150,000 agents by 2028, while only 13 percent of organizations believe their governance can handle that scale. The Discover and Contain features, which cover finding unregistered agents and sandboxing with a kill switch, remain in technical preview. The article also notes Salesforce's similar controls, with spending caps available now and a self-triggering kill switch delayed until January 2027.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Small RAG reranker experiment shows a tie, offering a verified pipeline but no proven gain

A developer fine-tuned a pretrained multilingual reranker and checked that its weights changed and that the saved checkpoint reloaded correctly. On a held-out synthetic slice of three questions, only two answerable, the adjusted and original rerankers both scored an nDCG@5 of 1.0, so the measured ranking gain was zero. The experiment isolated reranking by keeping the embedding model and the Qwen answer generator fixed, comparing four variants with and without retrieval and reranking. The work ran locally on CPU and is still in progress, with no real-domain fine-tuning, sealed real-data comparison, or human review of answers yet completed. The author concludes that the evidence supports a working experimental pipeline rather than any claim of improved answers.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Webhooks and polling together make crypto payment status tracking reliable

A crypto payment gateway confirms transactions asynchronously, since blockchain confirmations take time, so a backend learns of completed payments either by webhook or by polling. Webhooks are the fastest way to mark an order paid, but they can be delayed, retried, delivered out of order, or blocked, which risks paid customers with unpaid orders. A scheduled polling job that checks pending invoices acts as a slower, API-costly safety net for anything a webhook misses. The article's Express and Node.js example shows both paths calling one idempotent update so fulfilment happens only once. It also recommends verifying webhook signatures, responding quickly and queuing slow work, and logging each payment's source.

0 comments Read more at DEV Community