The Center for Investigative Reporting just sued OpenAI and Microsoft for copyright infringement, claiming the tech giants vacuumed up its stories to train AI models without permission or pay. It joins a mounting wave of publisher lawsuits like The New York Times, highlighting growing pushback against AI companies treating journalism as free training data.
OpenAI just introduced CriticGPT, a model based on GPT-4 designed to critique ChatGPT's code and help human trainers spot bugs. In experiments, trainers assisted by CriticGPT outperformed unassisted reviewers 60% of the time. The lab is now integrating these critic models into its RLHF pipeline to keep superhuman AI properly aligned.
Reverse-engineering group Rabbitude just revealed that Rabbit left critical API keys hardcoded in its codebase, allegedly allowing anyone to access every response the R1 has ever given. The keys could also let attackers brick devices or alter responses. Rabbit stated it was alerted on June 25 and has found no evidence of compromised user data so far.
The RIAA just filed massive copyright lawsuits against AI music generators Suno and Udio, claiming they unlawfully scraped vast catalogs of copyrighted songs to train their models and demanding up to $150,000 per infringing work. While the labels argue AI songalikes threaten their core business, the startups lean on fair use. Why it matters: this battle could redefine how all generative AI models license training data.
Anthropic just made Claude 3 Haiku and Sonnet available in AWS GovCloud and the AWS Marketplace for the US Intelligence Community. The startup also modified its usage policies to permit authorized foreign intelligence analysis like detecting sabotage and early military warnings, while strictly keeping bans on weapons and cyberwarfare.
Universal Music, Warner Music, and Sony Music just sued AI music startups Suno and Udio, alleging they illegally trained their models on copyrighted recordings without permission or payment. Led by the RIAA, the labels seek up to $150,000 per infringed work, totaling billions in potential damages. The copyright battles seen in text and images have officially hit AI music, likely forcing eventual licensing deals.
Apple just confirmed it won't roll out Apple Intelligence, iPhone Mirroring, or SharePlay screen sharing to EU users this year due to regulatory uncertainties from the Digital Markets Act. The company claims the DMA's strict interoperability rules could force compromises on user privacy and data security. Apple is working with the European Commission to find a workaround.
AI companies are quietly bypassing the decades-old robots.txt protocol to scrape publisher sites without permission, according to licensing startup TollBit. The finding follows Forbes recently calling out Perplexity for lifting its investigative work. With web scrapers ignoring traditional courtesy standards, publishers' primary defense against AI crawlers is essentially breaking down.
OpenAI just revealed updates on its Cybersecurity Grant Program, highlighting sponsored research across UC Berkeley, Dartmouth, and MIT after receiving over 600 applications. The projects focus on countering prompt injections, patching code vulnerabilities, and building autonomous defense agents. OpenAI is also expanding its support by granting free ChatGPT Team and Enterprise tiers to cybersecurity researchers.
Apple Intelligence will automatically add code to image metadata to flag AI alterations—even for simple fixes like removing a background distraction. Software chief Craig Federighi confirmed the policy, noting Apple has no plans to generate photorealistic people or places. Why it matters: As slop and deepfakes multiply, tech giants are leaning hard on provenance markers.
OpenAI co-founder and former chief scientist Ilya Sutskever just launched Safe Superintelligence Inc. (SSI), a new lab dedicated entirely to building safe superintelligence. The startup says it will insulate itself from short-term commercial pressures and product cycles, with zero plans to release anything else until safe superintelligence is solved.
Meta's FAIR team just released five new AI research models and tools to the public. Highlights include Chameleon for mixed text-and-image generation, JASCO for chord- and beat-controlled music synthesis, and AudioSeal, an audio watermarking system that detects synthetic speech up to 485 times faster.
U.S. cyber officials just teamed up with OpenAI, Microsoft, Nvidia, and AWS to run their first joint tabletop exercise simulating a cyberattack on a critical AI system. Over 50 experts gathered in Virginia to test incident response and communication channels, with findings set to shape an official AI security playbook due later this year.
OpenAI just appointed retired General Paul Nakasone, the former director of the NSA and US Cyber Command, to its board's Safety and Security Committee. While OpenAI claims his cybersecurity expertise will protect its AI supercomputers from hackers, whistleblower Edward Snowden blasted the move as a calculated betrayal and warned the public never to trust the company.
Meta just suspended plans to launch its Meta AI assistant in Europe following objections from the Irish Data Protection Commission over training on user data. Meta called the move a setback, arguing that blocking access to European content leaves it with only a "second-rate experience." Privacy advocates at Noyb countered that Meta simply refused to ask for explicit consent.
OpenAI just appointed retired U.S. Army General Paul M. Nakasone to its Board of Directors and Safety and Security Committee. The cybersecurity veteran and former head of U.S. Cyber Command will help guide the company in protecting critical model weights and supercomputing infrastructure from cyber threats. Why it matters: As AI systems grow more capable, securing frontier model infrastructure is becoming a paramount defense challenge.
A photographer just won an AI image competition using an actual photo of a flamingo. Miles Astray took third place and the People’s Choice award at the 1839 Color Photography Awards before judges realized it was real and disqualified him. Astray pulled the stunt to prove nature can still beat the machine.
Meta's AI is now injecting itself into Facebook comment sections to generate automated summaries, capturing everything from bizarre rants to lighthearted jokes. While the recaps often highlight just how chaotic discussions have become, the move is also reigniting privacy concerns over Meta training its models on user-generated content.
Google just detailed more than a dozen technical fixes to AI Overviews following viral blunders like suggesting users eat rocks or put glue on pizza. The company blamed data voids, satirical sites, and sarcastic forum posts for the hallucinations, and is now restricting user-generated content and humor from triggering AI answers.
California Governor Gavin Newsom just warned lawmakers that heavy-handed AI rules could stifle innovation and jeopardize the state's tech dominance. Speaking at the GenAI Summit, he noted 35 of the top 50 AI firms call California home, emphasizing that regulation is necessary but overreaching would put the state in a perilous position.
