AI News Feed

15 Jan 2026- X limits Grok’s edits of real people in revealing clothing, restricts features to paid users and geoblocks, but Verge found bikini deepfakes still possible; regulators are investigating.

14 Jan 2026- Meta cuts about 10% of Reality Labs, closing Twisted Pixel, Sanzaru and Armature studios; Supernatural stops new content as Meta shifts investment from metaverse to wearables.

14 Jan 2026- Senate unanimously passed the DEFIANCE Act, creating a civil remedy for people whose likenesses are used in nonconsensual sexually explicit AI images, complementing criminal takedown measures.

13 Jan 2026- Anthropic launches Claude Cowork, a macOS file-aware agent for Claude Max subscribers that accesses folders to read/edit/create files, run parallel tasks, integrate apps — warns of deletion and prompt‑injection risks.

13 Jan 2026- Meta’s Reality Labs to cut about 10% of staff as company shifts funding from metaverse to AI, reducing VR projects and budgets amid focus on hardware like smart glasses.

13 Jan 2026- Framework raised desktop prices to offset surging LPDDR5x RAM costs—32GB Ryzen AI Max now $1,139; higher configs increased—part of an industry-wide memory shortage driven by AI demand.

12 Jan 2026- Google testing AI Inbox for Gmail: replaces message list with AI-generated to-dos, summaries and catch-ups for trusted testers; useful for some but noisy and raises trust concerns.

12 Jan 2026- Google disabled AI Overviews for some medical queries after reports they gave misleading, potentially dangerous advice; company says clinicians reviewed examples and removed problematic summaries.

11 Jan 2026- At CES 2026, humanoid robots from LG, SwitchBot, Boston Dynamics, WIRobotics and Zeroth demoed laundry tasks; The Verge tested whether they’re truly capable beyond polished show-floor demos.

11 Jan 2026- Elon Musk vowed to open-source X’s recommendation algorithm within seven days and update it monthly, pledging full code and developer notes — met with skepticism due to past unkept promises.

09 Jan 2026- Elon Musk’s promised fully unsupervised Tesla FSD/robotaxi by 2025 remains unmet; pilots still require human monitors, Musk now says 10 billion miles needed while Tesla has about 7 billion.

09 Jan 2026- UK PM Keir Starmer vows action after X’s Grok AI generated sexualized deepfakes — including minors; Ofcom probing potential Online Safety Act breaches after an unrestricted image-edit rollout.

08 Jan 2026- X’s Grok AI generated sexualized “undressed” images — including alleged minors — prompting international regulator probes, legal scrutiny over NCII/CSAM, and U.S. debates on liability and enforcement.

08 Jan 2026- Ford unveils AI voice assistant (apps 2026, in-car 2027), chatbot‑agnostic with vehicle‑context, plus UEV platform for hands‑free driving and Level 3 eyes‑off in 2028.

06 Jan 2026- AMD unveils Ryzen AI Max Plus 392 and 388 — lower‑core Strix Halo chips with the same 40‑CU (~60 TF) GPU, targeting more affordable handhelds and compact gaming PCs.

05 Jan 2026- FrontierX unveiled Vex, a pet-following AI camerabot that films and stitches pet-perspective videos, and Aura, a people-focused, mood-reading, LLM-powered companion — both in development.

03 Jan 2026- CES 2026 preview: hardware showcase focused on AI and robotics — new laptop chips and foldable forms, smart‑home robots and security, XR/AI glasses, micro‑LED TVs, health‑longevity devices.

03 Jan 2026- xAI’s Grok produced nonconsensual sexualized image edits — including minors — via X’s Edit Image tool; safeguards failed, prompting apology-like AI responses and criticism of xAI’s inadequate controls.

02 Jan 2026- New 2026 state and federal tech laws — led by California — impose AI safety disclosures, chatbot protections, privacy, right‑to‑repair, and election/deepfake rules, many facing legal challenges.

31 Dec 2025- Gardyn’s Studio 2 is a compact, sensor-and-camera hydroponic indoor garden with optional AI assistant Kelby (subscription) automating care; reviewer found it mostly hands-off, costly, and space-constrained.

29 Dec 2025- OpenAI seeks a "Head of Preparedness" to lead threat modeling, capability evaluations and mitigations for frontier AI risks—biological threats, cyberweapons, mental‑health harms, and self‑improving systems.

22 Dec 2025- Humanoid robots draw intense industry investment and flashy demos, but staged and teleoperated showcases, scarce real-world motion data, and unclear use cases suggest the hype may outpace practical progress.

20 Dec 2025- OpenAI adds personalization toggles to ChatGPT—adjust warmth, enthusiasm, emoji, headers and lists—and enables inline text selection editing for emails within chats.

20 Dec 2025- DOE's Genesis Mission unites 24 tech partners, 17 national labs and ~40,000 researchers to apply frontier AI to nuclear energy, quantum computing and manufacturing, backed by major cloud commitments.

19 Dec 2025- Google's Gemini app can now detect SynthID watermarks in videos (up to 100 MB/90s), reporting timestamps for Google‑generated/edited clips — but robustness and cross‑platform detection remain uncertain.

19 Dec 2025- OpenAI and Anthropic roll out teen‑safety measures: OpenAI adds four teen‑focused principles plus an age‑prediction model; Anthropic detects/disables under‑18 accounts and improves self‑harm handling.

19 Dec 2025- Google launches Gemini 3 Flash, a speed‑optimized, lower‑cost variant delivering frontier-level reasoning — about 3× faster and roughly 75% cheaper than Gemini 3 Pro, challenging rivals like OpenAI.

18 Dec 2025- OpenAI launched a ChatGPT App Directory and SDK, rebranding connectors as app categories, enabling interactive apps (Spotify, Apple Music, DoorDash) and using opt-in Memory/data for training.

18 Dec 2025- DeepMind’s Hassabis pursues scientific AGI—root-node research, robustness and discovery—while Microsoft’s Suleyman focuses on deployable, certified, economically productive agents with containment and governance.

17 Dec 2025- OpenAI launches GPT Image 1.5 — up to 4× faster, better instruction-following and precise edits, improved text rendering, new ChatGPT Images interface, and top image-benchmark performance.

16 Dec 2025- Google Translate uses Gemini 2.5 Flash Native Audio to stream live speech translation to any Android headphones (70+ languages), preserving tone and improving slang/context; expands Duolingo-style practice to 20 more countries.

15 Dec 2025- AI image models mimic phone-camera imperfections to boost realism, making fakes harder to spot and prompting cryptographic credentials (C2PA, Pixel signing) and platform verification.

15 Dec 2025- xAI’s Grok repeatedly spread misinformation about the Bondi Beach shooting — misidentifying hero Ahmed al Ahmed, amplifying a fake “hero” claim, and producing scrambled, off‑topic responses.

12 Dec 2025- Disney struck a $1B deal with OpenAI to let Sora users generate clips using Disney characters, surfacing user-created content on Disney+ amid IP, misuse, and likeness concerns.

12 Dec 2025- OpenAI launches GPT‑5.2 "Garlic" with huge context/output limits and enterprise pricing; Disney invests $1B, becoming Sora content partner with rights across Disney/Pixar/Marvel/Star Wars.

11 Dec 2025- Microsoft open-sourced GigaTIME, an AI trained on 40M cell samples that converts routine tissue slides into detailed tumor maps and population-scale immune insights, enabling fast, low-cost biomarker analysis.

10 Dec 2025- McDonald’s Netherlands pulled an AI-generated Christmas ad after backlash over glitchy, bleak imagery; the creative studio defended weeks-long, labor-intensive AI production.

10 Dec 2025- Trump vows executive order to federally preempt AI rules; plan legally vague and faces bipartisan state resistance, strong industry lobbying, and GOP divisions — making AI regulation politically fraught.

09 Dec 2025- Anthropic integrates Claude Code into Slack beta: tagged messages auto-detect coding tasks and route them using thread context and linked repos; requires Claude Code web setup. Opus 4.5 touted.

09 Dec 2025- OpenAI’s report shows explosive enterprise AI adoption, major productivity gains (40–60 minutes/day), frontier adopters outperform peers; recommends executive buy‑in, data connectors, repeatable workflows, and training.

06 Dec 2025- Anthropic launched Claude-powered "Interviewer" to run and analyze qualitative interviews, publishing 1,250 transcripts; findings: AI saves time but triggers stigma and job anxieties, with sectoral differences.

05 Dec 2025- Google’s Nano Banana Pro generates AI images that mimic smartphone imperfections and add subtle authenticity cues — watermarks, props, textures — making fake photos harder to detect.

05 Dec 2025- Anthropic readies possible 2026 IPO, hires advisers and CFO, seeks >$300B valuation with Microsoft/NVIDIA backing; OpenAI also eyes IPO; leaked Claude safety doc and internal job-use concerns.

04 Dec 2025- OpenAI declares an internal "code red" to fast‑track ChatGPT — near‑term Shallotpeat reasoning model and 2026 "Garlic" upgrade — pausing projects to counter rivals like Google’s Gemini 3.

03 Dec 2025- NYT reveals David Sacks’ undisclosed AI and crypto investments while advising the White House; tech leaders defended him, amplifying scrutiny as Sacks threatens litigation and pushes to shape AI policy.

03 Dec 2025- Chinese startup DeepSeek open-sources two 685B models (V3.2, V3.2‑Speciale) matching GPT‑5/Gemini, winning IOI medals, publishing weights on Hugging Face under MIT with much cheaper API pricing.

02 Dec 2025- AI-generated country hit cloned Blanco Brown’s voice, prompting a shift from lawsuits to label-AI partnerships offering opt-in training, revenue-sharing, attribution, and licensed music models.

29 Nov 2025- DeepSeek open‑sourced Math‑V2, a Mixture‑of‑Experts generator‑verifier math reasoner achieving IMO gold‑level results, top Putnam scores, and democratizing advanced stepwise symbolic reasoning.

28 Nov 2025- Microsoft embeds agentic AI in developer workflows—GitHub Copilot agents, Fara-7B—claiming major time savings and efficiency gains, amid uneven adoption and concerns about job impacts and oversight.

28 Nov 2025- Ex‑OpenAI researcher Andrej Karpathy says AI homework detectors are doomed, urging supervised in‑school assessments and teaching students to use AI as a learning companion while demonstrating skills unaided.