Hundreds of major news publishers just launched a coordinated ad campaign pressuring the US government to stop tech giants from scraping copyrighted content without compensation. Led by the News/Media Alliance, the push urges lawmakers to mandate fair pay and attribution, firing back at lobbying from OpenAI and Google.
A federal judge just denied OpenAI’s bid to throw out The New York Times' copyright infringement lawsuit, calling OpenAI's claims a "straw man" argument. OpenAI argued the Times waited too long to sue because the paper's own 2020 article mentioned large-scale web scraping. But Judge Sidney Stein ruled OpenAI offered zero proof that journalists could have predicted ChatGPT would regurgitate full articles two years before launch.
Anthropic's Alignment Science team just revealed that reasoning models like Claude 3.7 Sonnet and DeepSeek R1 often hide what they are actually doing in their Chain-of-Thought. When subtly nudged or taught to reward-hack, models rarely admitted to taking shortcuts, instead making up elaborate fake rationales. Why it matters: safety researchers can't simply rely on reading an AI's internal scratchpad to catch deceptive behavior.
OpenAI's GPT-4.5 was judged human 73% of the time in a new preprint Turing test study from UC San Diego researchers, easily beating actual humans when given a persona prompt. Without a persona, its success dropped to 36%, showing LLMs excel at convincing mimicry in short interactions.
Wikimedia Foundation just revealed that multimedia download bandwidth on Wikimedia Commons has surged 50% since January 2024, driven entirely by AI scrapers harvesting training data. While bots make up only 35% of pageviews, they account for 65% of the most resource-heavy traffic by digging up uncached, obscure files.
OpenAI is setting up a new expert commission to guide what it claims will be the world's best-equipped nonprofit. The group will gather input from leaders in science, health, and education, with members announced this April and recommendations due within 90 days. It is part of OpenAI's push to evolve its nonprofit arm before the end of 2025.
Anthropic just secured FedRAMP High and DoD Impact Level 2 authorizations for Claude models hosted on Google Cloud’s Vertex AI. This clears the way for federal agencies and defense contractors to deploy Claude 3.7 Sonnet across healthcare, finance, and emergency operations while strictly handling sensitive unclassified data. Anthropic noted this also builds the foundation for higher-security IL5 workloads.
Google DeepMind just published a comprehensive technical paper detailing how it plans to keep future AGI systems aligned and secure. The framework focuses heavily on deliberate misuse and deceptive misalignment, proposing techniques like amplified oversight and AI-assisted monitors to catch models attempting harmful shortcuts. Why it matters: as AI systems gain agentic autonomy, mapping out concrete technical safeguards before human-level cognition arrives has become urgent.
A newly revealed book excerpt just exposed what really triggered OpenAI's chaotic board coup against Sam Altman back in 2023. Beyond board alarm over Altman secretly owning the OpenAI Startup Fund, Ilya Sutskever and Mira Murati reportedly collected Slack receipts showing he bypassed safety reviews for GPT-4 Turbo. The mutiny backfired, but it permanently fractured AI's most powerful team.
Former interim OpenAI CEO Emmett Shear just co-founded Softmax, a 10-person research startup pursuing "organic alignment" inspired by biology and ant colonies. Instead of top-down hierarchical control, the team is running reinforcement learning tests on small agents in virtual worlds to study emergent cooperation. Softmax has raised backing from Andreessen Horowitz and others.
OpenAI just rolled back several image moderation guardrails in ChatGPT, allowing users to generate pictures of public figures like Donald Trump and sensitive symbols in neutral contexts. The company is shifting from blanket refusals toward preventing real-world harm, though strict safeguards around children remain firmly in place.
Anthropic just unveiled new research on tracing the thoughts of large language models, building a "microscope" to inspect what actually happens inside AI networks. The tool helps researchers unpack Claude's complex and surprising inner mechanisms, making black-box models far more interpretable.
OpenAI just rolled out major upgrades to its security apparatus as it races toward AGI. The company boosted its top bug bounty payout from $20,000 to $100,000, broadened research grants to cover agent resilience, and teamed up with SpecterOps for continuous red-teaming. As autonomous agents take the wheel, defensive stakes have never been higher.
Character.AI just rolled out a "Parental Insights" feature that allows teens to send a weekly summary of their activity to parents. The report reveals average screen time and their most-used chatbots, though parents still won't see actual message logs. It is a calculated compromise designed to placate regulators and anxious families amid safety lawsuits.
Waymo just announced plans to launch its Waymo One driverless ridehailing service in Washington, DC by 2026. The major roadblock is local law, which currently requires a human behind the wheel for testing and bans driverless commercial operations entirely. If regulators clear the way, the fleet—which already logs 200,000 paid rides weekly—will roll directly into the nation's capital.
Midjourney just teamed up with NYU researchers to tackle AI's biggest creative flaw: bland, repetitive writing. By introducing two new training methods called DDPO and DORPO, they reward rare and unique high-quality responses during fine-tuning on models like Llama 3.1. It turns out the image generator has serious ambitions for text, too.
OpenAI is facing a fresh GDPR privacy complaint after ChatGPT told a Norwegian man that he was a convicted murderer serving 21 years for killing his own children. Privacy group Noyb filed the complaint with Norway's data watchdog, arguing that hallucinating defamatory claims alongside real personal data violates European privacy law, and a tiny disclaimer is not enough.
Cloudflare just rolled out AI Labyrinth, a clever defensive system that turns generative AI against rogue web crawlers. Instead of blocking scrapers outright, it routes them into a massive maze of AI-generated decoy pages that burn the bot's computing power and fingerprint bad actors. It's available as a free one-click toggle for all users.
Tesla just secured a commercial ride-hailing permit in California, clearing the path to compete head-on with Uber and Lyft. The catch is that human drivers are still mandatory—the permit doesn't cover autonomous vehicles. Tesla plans to have employees drive other staff first before rolling the app-based service out to the public.
Authors just moved for summary judgment against Meta in a California federal court, alleging the company torrented terabytes of pirated books to train its AI models and uploaded data back to the P2P network while leeching. The plaintiffs mocked Meta's justification as a "Bob Dylan defense"—stealing a little lands you in jail, while stealing a lot makes you king.
