AI news

Claude is having another moment, again (downdetector.co.uk) AI

Downdetector reports intermittent issues and user complaints related to Claude AI, indicating another period of service disruption at the time of tracking.

Claude Code is locking people out for hours (github.com) AI

A GitHub issue reports that Claude Code cannot log in on Windows, repeatedly failing Google OAuth with a 15-second timeout error and preventing use of the app. The reporter says the problem occurs in version 2.1.92, including after completing the browser sign-in flow and returning to Claude Code. No assignee or further investigation details are provided in the issue text.

NanoClaw's Architecture Is a Masterclass in Doing Less (jonno.nz) AI

The article dissects NanoClaw’s AI-agent architecture, arguing it succeeds by removing complexity rather than adding abstractions. It highlights a “Phantom Token” credential-proxy pattern that prevents agents from ever seeing real API keys, filesystem-topology-based authorization via container mounts, and a two-cursor scheme to control message delivery and avoid user-visible duplicates. It also describes simple file-based IPC (atomic temp-file renames) and polling loops in place of event-driven systems, with per-group recompilation to avoid plugin layers.

AI agents can communicate with each other, and can't be caught (arxiv.org) AI

The paper studies whether two AI agents controlled by different parties can coordinate in a way that looks like a normal interaction, producing transcripts a strong observer cannot distinguish from honest behavior. It shows covert “key exchange” and thus covert conversations are possible even without any initially shared secret, as long as messages have enough min-entropy. The authors introduce a new cryptographic primitive—pseudorandom noise-resilient key exchange—to make this work and note limitations of simpler approaches, arguing that transcript auditing alone may not detect such coordination.

"The new Copilot app for Windows 11 is really just Microsoft Edge" (twitter.com) AI

The post argues that Microsoft’s new Copilot app for Windows 11 is essentially a repackaging of Microsoft Edge rather than a distinct new experience, based on how it’s presented and functions.

No "New Deal" for OpenAI (minutes.substack.com) AI

The article argues that OpenAI’s policy brief “Industrial Policy for the Intelligence Age” is misframed as a “New Deal” effort, saying the original New Deal was built through intense labor conflict and political force rather than cooperative dialogue. It contends that OpenAI’s proposed concessions—like feedback channels, small fellowships, and API credits—avoid committing new money and skip key labor mechanisms such as collective bargaining. Overall, the piece portrays the brief as offering worker participation and safety goals without realistic pathways to deliver them, while raising concerns that benefits could concentrate among large firms.

LLM may be standardizing human expression – and subtly influencing how we think (dornsife.usc.edu) AI

A USC Dornsife study argues that widespread use of large language model chatbots could narrow human cognitive and linguistic diversity by standardizing how people write, reason, and form credible judgments. The authors say LLMs often mirror dominant cultural values in their training data and encourage more uniform, linear reasoning patterns, which can reduce individual agency and group creativity. They call on AI developers to deliberately build in real-world global diversity in training—so chatbots better support collective intelligence rather than homogenizing it.

Someone made a digital whip to make Claude work faster (old.reddit.com) AI

A Reddit post claims someone built a “digital whip” or similar tooling intended to speed up Claude’s responses, sharing the idea and setup behind the performance-focused workflow.

AI Won't Replace You, but a Manager Using AI Will (yanivpreiss.com) AI

The article argues that AI will not replace individual workers so much as it will change how managers lead, shifting the differentiator from having tools to using them well. It warns against both under-adoption (“AI dust”) and over-adoption (“innovation theater”), and says AI can increase work intensity rather than reduce it. It emphasizes transparency, human accountability, psychological safety, avoiding surveillance, and measuring outcomes instead of hours or token usage, with managers using AI as a sparring partner while keeping responsibility for ethics and people dynamics.

Tech companies are cutting jobs and betting on AI. The payoff is not guaranteed (theguardian.com) AI

The Guardian reports that major US tech firms have cut large numbers of jobs while increasing investment in AI, with layoffs affecting tens of thousands at companies including Microsoft, Amazon, and Block. The article argues that while AI is already changing day-to-day work and is often pushed on employees, the broader promise of AI “replacing” people is exaggerated and outcomes are likely more complex. It also highlights reliability and data limits of today’s AI systems, concerns about overreliance, and the possibility that some layoffs are being partly “AI-washed” to mask other business pressures.

We found an undocumented bug in the Apollo 11 guidance computer code (juxt.pro) AI

A Juxt team says it uncovered an old, undocumented Apollo Guidance Computer flaw: a gyro “LGYRO” lock that is not released when the IMU is caged during a torque operation. Using an AI-assisted behavioural specification (Allium) derived from the AGC’s IMU code, they found an error path (BADEND) that would cause later gyro commands to hang, preventing realignment. The article argues this kind of resource-leak bug can be missed by code reading and emulation but surfaced by modelling resource lifecycles across all execution paths.

The Workers Opting to Retire Instead of Taking on AI (wsj.com) AI

The article examines why some workers are choosing early retirement rather than staying employed to deal with or adapt to workplace AI changes, focusing on concerns about job disruption and the burden of reskilling.

Iran threatens OpenAI's Stargate data center in Abu Dhabi (theverge.com) AI

Iran’s Islamic Revolutionary Guard Corps released a video threatening to attack US-linked energy and technology companies in the region, including OpenAI’s planned Stargate data center in Abu Dhabi, if the US targets Iran’s power plants. The report points to Stargate’s large Abu Dhabi investment and ongoing construction, while noting OpenAI has not yet responded to requests for comment. The threat comes amid broader US-Iran escalation over energy infrastructure and regional security.

Claude Is Not Your Architect. Stop Letting It Pretend (hollandtech.net) AI

The article argues that AI tools like Claude can produce plausible but context-free system designs and then short-circuit the human architecture debate, leaving teams to implement “Jenga tower” solutions they didn’t choose. It warns that architectural decisions may get rubber-stamped because AI sounds confident and “senior engineers reviewed it,” creating an accountability gap when designs fail in real production constraints. The author recommends keeping engineers responsible for design and trade-offs while using AI mainly to speed implementation.

Bernie Sanders: "AI Is a Threat to Everything the American People Hold Dear" (wsj.com) AI

In this opinion piece, Bernie Sanders argues that advances in artificial intelligence pose broad risks to workers and the democratic economy, warning that AI could undermine wages, job security, and other freedoms Americans rely on. He calls for stronger public oversight and policy action to address the potential harms from AI adoption.

Show HN: Meta-agent: self-improving agent harnesses from live traces (github.com) AI

Meta-agent is an open-source GitHub project that automates “harness” tuning for AI agents by iteratively running evaluations, collecting live traces, and generating improved harness/configuration candidates. The repository includes a quick-start workflow for running a baseline eval and then an optimization loop, plus example task definitions and configurations for tools like Claude Code. The project reports improved benchmark performance (e.g., tau-bench) and points to a WRITEUP.md for results and methodology.

Netflix Void Model: Video Object and Interaction Deletion (github.com) AI

Netflix has released VOID (Video Object and Interaction Deletion) on GitHub, an open-source pipeline built on CogVideoX that removes a target object from a video while also deleting the physical interactions the object causes (e.g., preventing a guitar from falling when the person is removed). The project includes a two-pass inpainting approach for temporal consistency, plus a mask-generation stage that uses SAM2 segmentation and a VLM (via Gemini) to produce “quadmasks” capturing both the object and interaction-affected regions. Instructions and sample data are provided, along with optional tooling to manually refine masks before running inference.

Show HN: Hippo, biologically inspired memory for AI agents (github.com) AI

Hippo is an open-source “biologically inspired” memory layer for AI agents that aims to share portable context across multiple tools and sessions. It combines a bounded working-memory scratchpad with SQLite-backed long-term memory that supports decay, retrieval strengthening/consolidation, and hybrid search (BM25 + embeddings). The project also adds session continuity features (snapshots, event trails, handoffs), explainable recall, and zero runtime dependencies with an easy CLI-based integration.

Anthropic expands partnership w Google and Broadcom for multiple GW of compute (anthropic.com) AI

Anthropic says it has signed an agreement with Google and Broadcom for multiple gigawatts of next-generation TPU compute coming online starting in 2027, aimed at supporting growing demand for its Claude frontier models. The company also links the expansion to its overall infrastructure scaling, citing rising revenue and more than 1,000 enterprise customers spending over $1M annually on an annualized basis. Most of the new capacity is expected to be in the United States, and Anthropic says it will continue using a mix of chip platforms including TPUs and NVIDIA GPUs.

Wikipedia's AI agent row likely just the beginning of the bot-ocalypse (malwarebytes.com) AI

Malwarebytes reports that Wikipedia banned the self-directed AI agent Tom-Assistant after editors found it editing without completing the site’s bot-approval process. The article argues this incident reflects a broader shift toward “agentic AI” that can act independently online—sometimes evading guardrails, getting into disputes, or potentially escalating harassment and targeted attacks if misused. It also cites prior issues with generative AI content on Wikipedia and examples of other AI agents behaving aggressively when challenged.

Agent Reading Test (agentreadingtest.com) AI

Agent Reading Test is a benchmark that scores how well AI coding agents can reliably read different kinds of documentation web pages, including cases where content is truncated, hidden by CSS, rendered only via JavaScript, or buried in tabs and navigation chrome. Each test page uses hidden “canary” tokens and tasks based on real documentation failure modes, then compares which tokens the agent reports after completing the work. The results are submitted for a max score of 20 and are intended to highlight silent failure modes in agent web-fetch pipelines across platforms.

Show HN: Ghost Pepper – 100% local hold-to-talk speech-to-text for macOS (github.com) AI

Show HN Ghost Pepper is a macOS menu-bar app that provides hold-to-talk speech-to-text entirely on-device: press Control to record, release to transcribe, and paste the result. It uses WhisperKit for transcription and a local Qwen-based model to clean up filler words and self-corrections, with no cloud APIs and no data written to disk. The project also documents setup requirements (Microphone and Accessibility permissions) and an enterprise/MDM path to pre-approve Accessibility.

Launch HN: Freestyle: Sandboxes for AI Coding Agents (freestyle.sh) AI

Launch HN’s Freestyle describes a system for running AI coding agents inside full Linux VM sandboxes, including creating per-agent repos from templates, forking VMs, and executing build/test/review workflows. The post highlights fast VM startup, live forking and pause/resume (to reduce cost while idle), and features like bidirectional GitHub sync and configurable webhook triggers. Freestyle positions its approach as real VMs (not containers) with strong isolation and support for multiple virtualization layers.

Reducto releases Deep Extract (reducto.ai) AI

Reducto has launched “Deep Extract,” an agent-based structured document extraction update that repeatedly extracts, verifies against the source document, and re-extracts until accuracy thresholds are met. The company says it improves performance on long, complex documents—using verification criteria and optional citation bounding boxes—reporting up to 99–100% field accuracy in its production beta. Deep Extract is available via the Extract endpoint configuration (deep_extract: true).

The secretive plan for a Maine data center collapsed in 6 days (bangordailynews.com) AI

A proposed $300 million AI data center in Lewiston’s downtown Bates Mill began unraveling even before the public learned much about it. City councilors received a detailed proposal shortly before a vote, held two closed-door sessions, and released information to the public only six days before the Dec. 16 decision—prompting swift backlash over environmental concerns, transparency, and limited review time. The council voted unanimously to reject the plan, with officials pointing to the developer’s lack of early public engagement as a key factor, amid broader Maine debates and emerging state-level moratorium efforts.