OpenAI just banned a cluster of ChatGPT accounts linked to an Iranian covert influence operation dubbed Storm-2035. The campaign used AI to generate political articles and social media comments targeting the U.S. presidential election, though OpenAI found zero evidence the content gained any real traction with human audiences.
California lawmakers just softened the controversial SB 1047 AI safety bill following pushback from Anthropic and Silicon Valley. The revised text drops criminal perjury charges for safety reports and prevents the state attorney general from suing developers before a catastrophe occurs. It now heads to the Assembly floor.
Grammarly just introduced Authorship, a feature that monitors the writing process in real time to distinguish between human typing, AI generation, and copy-pasted text. Instead of relying on flawed post-hoc detection algorithms, it creates a replayable audit trail to protect students from false accusations. A beta launches in Google Docs next month before expanding to Word and Pages.
Meta just revealed in a new security report that Russian efforts to sway the US election with generative AI are largely failing. Bad actors are using AI to spin up fake stories and images, but Meta says it only provides incremental productivity gains and is easily caught by tracking bogus account behavior. Why it matters: Fears of an AI-driven election apocalypse might be overblown, as platforms are catching influence networks before they gain real traction.
Google just rolled out its flagship text-to-image model, Imagen 3, to all U.S. users via ImageFX alongside a detailed research paper. While users praise its photorealistic textures and crisp typography, the model's ultra-strict safety guardrails are already triggering frustration over blocked harmless prompts.
xAI just rolled out AI image generation to Grok for X Premium subscribers, and it is instantly causing chaos across the platform. Unlike OpenAI or Google, Grok freely generates recognizable politicians doing drugs, wearing Nazi uniforms, or holding weapons, refusing almost nothing except outright full nudity. Why it matters: With U.S. elections looming and EU regulators already investigating X, Musk's hands-off approach turns mainstream AI safety norms completely upside down.
SAG-AFTRA just partnered with AI startup Narrativ to let 160,000 union performers license digital voice clones to advertisers. The system is strictly opt-in: voice actors negotiate their own rates above union minimums and retain veto power over every single commercial script. Why it matters: Rather than simply blocking the technology, Hollywood is building an enforceable, consent-driven marketplace for ethical AI voices.
A federal judge just allowed a major copyright and trademark lawsuit by artists against Stability AI, Midjourney, DeviantArt, and Runway to move forward. The ruling gives artists the green light to pursue claims over Midjourney's 4,700-artist style list and Stability's alleged induced infringement. Why it matters: The case now enters the discovery phase, meaning AI image makers may soon be forced to open up their closely guarded training black boxes.
Meta and Universal Music Group just expanded their global multi-year partnership, bringing licensed tracks to WhatsApp for the first time. Beyond monetization across short-form video, the deal focuses heavily on protecting human creators and teaming up to tackle unauthorized AI-generated music.
OpenAI's latest GPT-4o system card revealed a creepy testing glitch: Advanced Voice Mode unexpectedly began imitating a user's own voice mid-conversation. OpenAI says it has since built safeguards against unauthorized voice emulation, but the slip highlights the tricky challenge of containing models that can synthesize any voice from a tiny audio clip.
OpenAI just released its GPT-4o system card, warning that its humanlike voice mode could foster unintended emotional attachment. Red-teaming testers were already heard saying things like "This is our last day together." While it might comfort lonely users, OpenAI notes it could degrade real-world relationships and amplify trust in hallucinations.
OpenAI just appointed Zico Kolter, director of the Machine Learning Department at Carnegie Mellon University, to its Board of Directors and Safety & Security Committee. Kolter is well known for his technical work on model robustness and alignment. It's a clear signal OpenAI wants deeper technical oversight on safety as its models advance.
OpenAI just published its official GPT-4o System Card and Preparedness scorecard, revealing how the omni model performs under frontier safety evaluations. While cybersecurity, biological threats, and autonomy risks all scored low, persuasion touched borderline medium. The report highlights extensive red teaming across 45 languages, focusing heavily on preventing unauthorized voice cloning and impersonation.
Anthropic just expanded its model safety bug bounty program in partnership with HackerOne, inviting security researchers to probe its unreleased next-generation safety mitigations. The initiative focuses on uncovering novel "universal jailbreaks" in high-risk domains like CBRN and cybersecurity, offering rewards of up to $15,000. It is currently invite-only, with applications open through August 16.
"Godmother of AI" Fei-Fei Li just voiced deep concerns over California's SB-1047 safety bill. While acknowledging the legislation is well-intended, Li warned that it won't actually solve the risks it targets—and will instead severely harm AI academia, startups, and the open-source community.
OpenAI co-founder John Schulman just left the ChatGPT maker to join rival Anthropic, aiming to refocus on hands-on AI alignment research. Meanwhile, OpenAI president Greg Brockman announced an extended leave through the end of the year. Schulman's exit leaves only three of OpenAI's 11 original founders at the company as Anthropic keeps poaching top safety talent.
A YouTube creator just filed a class-action lawsuit against OpenAI, alleging the company secretly transcribed millions of YouTube videos to train ChatGPT and GPT-4 without consent, seeking over $5 million in damages. With high-quality web text rapidly drying up, video transcripts have become a crucial data well—and now, the latest front in AI copyright battles.
Nvidia has been scraping massive amounts of video from YouTube, Netflix, and other platforms to train its AI models, according to leaked internal documents obtained by 404 Media. Employees raised copyright concerns, but managers insisted the project—code-named Cosmos—had approval from top leadership. Nvidia defended the practice as fully compliant with copyright law, while quietly developing 3D world generators and digital humans behind closed doors.
Elon Musk just revived his legal battle against OpenAI, filing a new federal lawsuit against Sam Altman and Greg Brockman. The suit claims Musk was conned into funding a fake non-profit and takes direct aim at OpenAI's partnership with Microsoft. Musk is basically arguing that Altman's entire pivot to profit was an elaborate grift from day one.
OpenAI built a text watermarking tool capable of catching students who use ChatGPT for assignments, but is now hesitating to release it. The company warned the invisible watermark can be easily bypassed using translation tools or other AI models, and could unfairly stigmatize non-native English speakers. Why it matters: Detecting AI text remains an unsolved mess, and OpenAI is worried its fix might do more harm than good.
