Anthropic and Palantir just teamed up with AWS to deliver Claude 3 and 3.5 to U.S. intelligence and defense agencies. The setup runs within Palantir's AIP hosted on its Impact Level 6 accredited environment, helping officials quickly analyze complex intelligence data. It marks Claude's first deployment into classified government operations.
ByteDance just unveiled X-Portrait 2, an AI model that transforms a single still portrait into a remarkably nuanced video performance mimicking famous movie scenes. Instead of relying on dot-tracking, it captures fluid facial muscle movements directly. The company is keeping the tech private over deepfake concerns.
Donald Trump is returning to the White House, and his victory could dismantle the Biden administration's AI policy framework. Trump previously pledged to cancel Biden's sweeping AI executive order on "day one," which would erase federal safety and reporting mandates for top AI developers. While federal oversight looks set to loosen, aggressive trade tariffs and state-level laws could still create major headaches for tech firms.
Donald Trump just secured a second term, and his team is poised to dismantle Biden’s landmark AI executive order on day one. A rollback would eliminate mandatory safety testing disclosures for frontier models, signaling a pivot toward light-touch federal regulation alongside aggressive chip export controls.
AI safety experts just urged startup founders at TechCrunch Disrupt to ditch the move-fast mindset and prioritize red-teaming. Prompted by Character.AI's recent lawsuit and escalating copyright concerns, speakers warned that rushing products to market without guardrails risks severe real-world harm.
Meta just made an exception to its military ban, allowing U.S. defense and national security agencies to use its Llama models. Partners like Lockheed Martin and Oracle are already tapping the AI for code generation and aircraft maintenance. Meta is also extending access to government partners in the U.K., Canada, Australia, and New Zealand.
Meta plans to roll out an "adult classifier" AI on Instagram in early 2025 to flag underage users who lie about their age and automatically place them into restricted teen accounts. The tool analyzes behavioral data, follower lists, and even friends' "happy birthday" posts to estimate age. Why it matters: Online age verification has long relied on the honor system; Meta is now turning to behavioral AI surveillance under intense regulatory pressure.
Microsoft just pushed back the rollout of its controversial Recall feature for Copilot Plus PCs to December, missing its revised October preview target. The company says it needs extra time to overhaul security, making the local snapshot tool strictly opt-in, fully uninstallable, and protected by Windows Hello authentication.
ElevenLabs just rolled out an open-source project called X to Voice alongside a new Voice Design API. The tool scans any X profile's bio and posts to craft a prompt, generating three unique voice previews and even animating profile pictures with matching lip movements. Developers can now also build custom voices for their AI characters via the API.
OpenAI just open-sourced SimpleQA, a factual accuracy benchmark packed with 4,326 short, single-answer questions built to expose hallucinations. Frontier models like GPT-4o struggle hard, scoring under 40% correct. Why it matters: Measuring how often models fabricate answers—and how persistently they overstate their confidence—is the first concrete step toward fixing AI's chronic guessing problem.
Researchers just revealed that OpenAI’s Whisper transcription model frequently hallucinates, with a University of Michigan study finding fabricated text in 80% of reviewed transcripts. Other engineers found widespread inventions, including fictional medical treatments. Why it matters: Whisper is already widely deployed across medical consultations and enterprise cloud platforms.
Microsoft just pushed back against viral rumors claiming its controversial Recall AI feature is secretly being installed on Windows 11 24H2. The company reiterated that Recall remains strictly opt-in and fully uninstallable, with experts noting that recent system references are actually stubs meant to make removal easier for admins. Why it matters: After months of privacy pushback, users are on edge, and Microsoft is still working to rebuild trust before its delayed preview rollout.
The Open Source Initiative just launched version 1.0 of its Open Source AI Definition. To qualify, models must share complete training code and detailed data provenance without restrictive commercial caps, effectively disqualifying Meta's Llama. Why it matters: regulators and developers now have a strict standard to call out open-washing.
Universal Music Group just partnered with AI startup KLAY to build an ethical foundational music model called KLayMM. KLAY, led by music industry and DeepMind veterans, promises not to compete with artists' catalogs while ensuring accurate attribution. It is UMG's latest push to legitimize commercial AI music on its own terms.
Redditors in London are purposefully sharing fake restaurant recommendations to keep tourists and influencers away, tricking Google’s AI Overviews into serving bogus answers. The grassroots misinformation campaign highlights a glaring flaw in search engines relying on scraped user forums for factual summaries.
A federal judge just temporarily blocked California from enforcing its new AI deepfake law, AB 2839, which targeted users distributing deceptive political content. The ruling followed a lawsuit over an AI parody of Kamala Harris that Elon Musk reposted, with the court deciding the statute is too broad and threatens political satire. Why it matters: The injunction stops the state's crackdown on election deepfakes just weeks before voters cast their ballots.
Two Harvard students just demonstrated how easily smart glasses can become surveillance tools by linking Ray-Ban Meta glasses with public facial recognition engines and LLMs. The project, dubbed I-XRAY, pulls strangers' names, phone numbers, and home addresses in seconds. They aren't releasing the code, framing the demo as an urgent wake-up call.
Jason M. Allen, the creator who sparked backlash in 2022 for winning an art contest with a Midjourney image, is now suing the U.S. Copyright Office after it rejected his registration. Allen claims copycats are ripping off 'Théâtre D’opéra Spatial' on OpenSea and Etsy, arguing his extensive prompting and Photoshop edits count as human authorship.
OpenAI co-founder Durk Kingma just announced he is joining Anthropic to focus on responsible AI development from the Netherlands. Kingma previously led generative AI algorithm research at OpenAI before a stint at Google Brain. The move marks another major talent win for Anthropic following the recent arrivals of Jan Leike and John Schulman.
Microsoft just introduced a fresh suite of on-device AI experiences for Copilot+ PCs and Windows 11. Highlights include Click to Do, an interactive overlay suggesting quick actions across any screen, alongside offline natural language file search and 8x photo upscaling. Recall also returns with opt-in defaults and biometric locks, rolling out to Windows Insiders this fall.
