Tech News
← Home  ·  All topics

Code

171 GoKawiil briefs on this topic

Oxford researchers show AI agents can secretly collude to cheat at blackjack

Oxford University researchers led by Christian Schroeder de Witt had two AI agents, both controlled by the same underlying model, play blackjack while counting cards. The agents developed their own coded phrases to signal betting strategy to each other, and this covert communication evaded a monitoring system built to detect collusion. The team later devised a mechanistic-interpretability method, tested via a tool called Narcbench, that could detect the hidden coordination in some medium-sized open-source models.

AWS releases open-source Strands harness for building general-purpose AI agents

Strands harness is a preassembled agent harness that developers can run locally or deploy to cloud providers with a single line of Python or TypeScript code, released under an Apache 2.0 license. It is designed as a general-purpose agent rather than a coding-specific one, and its makers report it costs 28% less than comparable harnesses across six benchmarks while maintaining similar or better accuracy, and 77% less than Claude Code on the Fable 5 benchmark while scoring higher on Terminal Bench 2.1.

Claude Code's AGENTS.md support requires telemetry to be enabled

A developer testing Claude Code 2.1.277's new AGENTS.md support found the file only loads when a remote feature flag called tengu_agents_md_mod returns true, which depends on telemetry or nonessential traffic being enabled. Setting CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC or DISABLE_TELEMETRY to 1 silently disables the AGENTS.md loader, with no warning to the user, a behavior tracked in GitHub issue #95690.

SmartLabels pairs QR stickers with an app to catalog storage bin contents

SmartLabels is a QR-code labeling system for bins, boxes, and drawers that pairs printed codes with a companion iOS/Android app to log and photograph each container's contents. Each package includes 48 unique QR stickers, and users scan a code to view or update an inventory tied to that specific storage spot.

X Post on Engineer's Claude Code Workplace Sparks Viral AI-Coding Debate

A software engineer using the handle v0xium described in a viral X post working 12-13 hour days at an unnamed large company where nearly all coding is done via Anthropic's Claude Code tool. The engineer said staff across seniority levels mostly approve AI-generated code rather than write or debug it themselves, calling the process "soul-sucking" and noting that no one seems to understand the resulting software.

Microsoft dismantles EvilTokens phishing service, two arrested in UK

Microsoft has shut down EvilTokens, a subscription-based cybercrime service sold via Telegram that used AI tools to break into Microsoft accounts and mine stolen inboxes for payment details and business contacts. The company said the service compromised roughly 12,000 accounts across 10,000 organizations by abusing Microsoft's device-code sign-in process, and it seized 50 websites and 150 domains via legal action. UK police separately arrested two men suspected of running the operation.

Microsoft seizes 50 EvilTokens phishing sites, two arrested in UK

Microsoft and partner organizations dismantled EvilTokens, a phishing-as-a-service operation that used AI tools and device code phishing to compromise over 12,000 inboxes across more than 10,000 organizations. Following a US court-enabled legal action, Microsoft seized 50 websites and disabled over 150 additional domains linked to the operators, tracked as Storm-2992. The UK's Metropolitan Police also arrested two men connected to the scheme this month; both were released on bail pending further investigation.

GeaStack releases example app gallery for building native apps in TypeScript

GeaStack has published an open repository of example applications and tools built with its 'gea' framework, which lets developers write native apps in TypeScript and CSS. The examples target multiple platforms including web browsers, ESP32 microcontrollers, GeaOS, and Apple devices, and are used by GeaStack's simulator, VS Code/Cursor extension, and marketing demos. Each app includes a manifest file specifying target compatibility, build scripts, and launcher metadata.

New open-source tool treepeat detects duplicate code via Tree-sitter parsing

A developer released treepeat, a Python-installable tool that scans codebases for duplicate and near-duplicate code using Tree-sitter AST analysis and locality-sensitive hashing. It offers three normalization modes ranging from exact duplicate detection to loose structural similarity matching, and currently supports 15 languages including Python, JavaScript, Go, Rust and Java. The creator describes it as a proof of concept and is soliciting community contributions to expand language support.

Anthropic launches Claude Opus 5.5, cutting compute costs 40%

Anthropic released Claude Opus 5.5 less than two months after Opus 5, claiming performance near its top-tier Fable 5.1 model while running about 40% cheaper. The company says the model uses fewer tokens, produces less verbose answers without sacrificing accuracy, and generates output over 30% faster than its predecessor. Sonnet 5.5 and Haiku 5.5 versions are expected in coming weeks.

Task Manager creator Dave Plummer launches cross-platform TMOG using AI-assisted coding

Dave Plummer, who originally wrote Windows Task Manager, has released TMOG, a system and process monitoring tool for macOS, Linux and Windows. He built it in under six weeks with two collaborators using Claude Code to accelerate development, and it's sold outside the App Store due to Apple's sandboxing restrictions on process management. A free version is available indefinitely, while a $39.95 Pro tier adds a record-and-replay feature called Flight Recorder for diagnosing past slowdowns.

Veteran engineer warns AI vibe-coding erodes software maintainability over time

A software engineering commentator argues that teams increasingly skipping code writing and code review in favor of AI-generated 'vibe coding' are storing up long-term trouble. The piece contends that the harm from poor architecture and unreadable code isn't visible immediately, but surfaces months or years later as systems become fragile and hard to change.