AI Ethics news and analysis on Web Pulse: 16187 curated articles tracking the latest AI Ethics developments, tools, and research, updated continuously from vetted sources.
Twitch's Chief Product Officer admitted during a public exchange that the platform's new AI training feature is widely disliked, acknowledging that if it were opt-in, nobody would choose it, yet the company plans to proc…
His Majesty’s Courts and Tribunals Service (HMCTS) has banned Meta smart glasses from courts and tribunals in England and Wales, with devices confiscated at the door and returned when the owner leaves. The ban follows a …
Anthropic announced it is enabling watermarking for AI-generated outputs from its Claude model, a move that allows the company to trace the origin of text produced by the system. The watermarking is designed to help iden…
Gil Zilberfeld, a developer and tester, warns that AI-generated code introduces a new form of technical debt, which he calls 'AI code debt.' He argues that this debt is larger and more dangerous than traditional technica…
A blog post argues that 'agentic engineering' is simply the application of standard software engineering practices—such as using issue trackers, writing documentation, and maintaining tests—that teams often skip under ti…
Amazon is using Twitch's user-generated content to train its generative AI models, a move that has drawn criticism from the platform's users. The company confirmed that it will use channel content for AI training, raisin…
SpaceXAI released Grok 4.6 on August 12, 2026, matching GPT-5.6 Sol with a score of 61 on the Artificial Analysis Intelligence Index, but the model trails competitors on coding benchmarks, recording 26% on Terminal-Bench…
A Wall Street Journal investigation found that Chinese censorship is influencing answers from American AI systems, with models trained on Chinese data or aligned to Chinese regulations producing politically sensitive res…
Watermarks-remover v0.3.2, an open-source agent skill and Python script suite by Guillaume Meyer, strips multi-vendor AI provenance marks from text and files, including Claude, Gemini/SynthID-Text, OpenAI, and open-LLM K…
Alibaba Group Holding Ltd. released the Qwen3.8-2.4T-A95B open-weights model under a restrictive custom Qwen3.8-Max License, requiring businesses with over $50 million in annual revenue to negotiate a separate commercial…
A new reverse-engineering technique using inverse language models can reconstruct proprietary system prompts from generated outputs, exposing the 'secret sauce' of commercial LLM agents. This vulnerability enables intell…
Mozilla CEO Anthony Enzor-DeMeo said the organization is deliberately late to AI, focusing on giving users model choice in its Firefox browser and upcoming Smart Window feature rather than building frontier models. He em…
Engineering teams can build custom AI applications like RAG-based chatbots in a weekend, but securing them requires a new approach. The attack surface shifts from infrastructure to the AI workflow itself, with risks from…
Brian Roemmele, founder of technology research site ReadMultiplex, released a Mac-only tool and local rewriting workflow on August 12th that he says can remove the machine-readable watermarks Anthropic is building into C…
Enterprise governance research shows that every AI agent in production needs a named human owner, not a team, to close accountability gaps that lead to costly failures. Regulators and courts are already applying existing…
A survey of 300 tech executives reveals that AI agents on average access only 45% of a company's data, while 'data leaders' provide over 70% access and fully trust their agents, compared to only half of all organizations…
A Wikipedia editor revised the Flock Safety surveillance article to add misuse cases in Milwaukee and Georgia, correct a misquoted student, change a Columbine Valley recording to ring footage, and cite Kansas case number…
A new arXiv study (2608.11256v1) finds that commercial AI detectors cannot reliably distinguish AI-assisted editing from full AI drafts, flagging guideline-compliant 'refine abstract only' edits at 64–80% rates (Pangram/…
A new arXiv paper (arXiv:2608.11225v1) proposes a conceptual framework and research program for AI personality clones, defining identity through observer-assessed indiscernibility of manifestations and distinguishing fid…
A new arXiv study (2608.11390v1) finds that generative engine optimization (GEO) attacks can escalate into citation wars, degrading document quality and introducing unsupported claims, and proposes a platform–creator mec…