Claim: Claude Code Allegedly Embeds Steganographic Markers in Requests

·1 views

A blog post published on thereallo.dev alleges that Claude Code, Anthropic's AI coding assistant, is steganographically marking its requests in some form. The claim gained traction on Hacker News, accumulating 90 points and community discussion. Steganography involves hiding information within data in a way that is not immediately obvious to users. The full technical details and methodology behind the claim are outlined in the original blog post. Anthropic has not publicly responded to the allegation based on available information.

Read the full story at Hacker News

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

Discussion (0)

How to Detect AI Chat Platforms in Chrome Extensions Using URL and DOM Checks

Developers building Chrome extensions for AI chat platforms like ChatGPT, Claude, Gemini, and Copilot need a reliable way to identify which platform is active before executing any scripts. The recommended approach uses a two-tier detection system: first checking the page's hostname for a fast, near-instant match, then falling back to DOM signature patterns for edge cases like embedded interfaces or white-label deployments. DOM-based detection uses a majority-vote method, requiring at least two structural checks to pass, which reduces false positives on unrelated pages. Because AI chat platforms are single-page applications that load their UI asynchronously, extensions should inject content scripts at document_idle rather than document_start to ensure the DOM is ready. Together, hostname detection covers over 95% of real-world cases, with DOM-based fallback serving as a safety net for the remainder.

0 comments Read more at DEV Community

ProgrammingHacker News ·

Scammers Use AI-Generated Flower Images to Sell Fake Exotic Seeds Online

Fraudsters are selling seeds for exotic flowers that do not actually exist, using AI-generated images to make the listings appear convincing. The scam exploits the growing appeal of rare and unusual plants, luring buyers with visually striking but entirely fictional flower varieties. Customers who purchase the seeds either receive nothing, or are sent generic seeds that bear no resemblance to the advertised product. The scheme highlights how AI-generated imagery is increasingly being weaponized to deceive consumers in online marketplaces.

0 comments Read more at Hacker News

ProgrammingDEV Community ·

How to Split a Terraform Monolith into Smaller, Focused State Files

Large Terraform states can slow down plans, widen blast radius, and cause conflicts when multiple teams work simultaneously. The recommended approach is to group resources by lifecycle and ownership into logical boundaries such as networking, compute, application, and monitoring layers. Engineers can migrate resources between states without destroying them using the terraform state mv command, moving one logical group at a time and verifying plans after each step. Once split, cross-state dependencies must be rewired by replacing direct resource references with input variables and module outputs. Tools like remote state data sources or dependency management platforms can then be used to pass output values across the separate state files.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Why Large Terraform State Files Slow Teams Down and How to Fix Them

As Terraform projects grow, state files managing hundreds or thousands of resources cause plan times to balloon, with some users reporting 20–25 minutes for states containing around 2,900 resources. Every terraform plan refreshes all resources via API calls regardless of whether their configuration changed, creating a sequential bottleneck that worsens with each new resource added. Large shared states also trigger cloud API rate limits from providers like AWS, Azure, and GCP, sometimes causing plans to fail entirely due to throttling. Beyond performance, a single large state file creates a wide blast radius where unrelated resources can be accidentally modified or destroyed in the same apply. Splitting infrastructure into smaller, purpose-specific state files is identified as the structural solution, while proposals for partial-refresh features in Terraform and OpenTofu remain unimplemented.

0 comments Read more at DEV Community