Top Stories
Google Unveils Gemini 3.5, Spark Agent, and a $100/Month AI Ultra Tier
Google I/O 2026 delivered the goods yesterday. The headline: Gemini 3.5, which Google calls “a major leap forward in building more capable, intelligent agents.” The first model shipping is Gemini Omni Flash, rolling out now to the Gemini app, Google Flow, and YouTube Shorts.
The bigger deal for most people is Gemini Spark, a new AI agent that lives inside the Gemini app and can reason across your connected apps. Google is positioning it as an “active partner that does real work on your behalf” — not just a chatbot that answers questions but something that can take multi-step actions across apps, browse the web, fill out forms, and create custom widgets through natural language.
Spark is going to trusted testers this week, with a beta for U.S. Google AI Ultra subscribers next week. That Ultra tier is new too — $100 per month, aimed at developers, creators, and power users. Google also showed off Android XR glasses built with Samsung, Gemini-powered Android 17 features, and upgrades to Google Flow with new agentic tools and music-making capabilities.
Federal Deepfake Removal Law Now Has Teeth
The Take It Down Act hit its enforcement milestone yesterday, exactly one year after President Trump signed it into law. What changed: the FTC can now fine platforms $53,088 per violation for failing to remove non-consensual intimate images — including AI-generated deepfakes — within 48 hours of receiving a removal request.
Every covered platform (social media, messaging apps, image hosts, gaming platforms) was required to have a notice-and-removal process in place by May 19. Anyone depicted in a non-consensual intimate image can file a removal request, and the platform must take it down along with known copies within two days.
The timing is notable given the explosion of AI-generated deepfakes over the past year. Gizmodo flagged questions about how platforms like Grok will handle compliance, given xAI’s historically loose content moderation stance.
Source: TechTimes | Fisher Phillips
OpenAI Partners with Dell to Bring Codex On-Premises
OpenAI made its first explicit play for hybrid and on-premises enterprise deployment, announcing a partnership with Dell Technologies to run Codex on Dell’s AI Data Platform. More than 4 million developers now use Codex weekly, but until now it lived entirely in the cloud.
The partnership matters most for regulated industries — financial services, healthcare, government — that cannot send code and data to public cloud endpoints. Codex will connect with Dell’s AI Factory infrastructure for data preparation, system management, testing, and deployment. It is a clear signal that OpenAI sees enterprise on-prem as a major growth vector, not just a concession to compliance teams.
Source: OpenAI
Quick Hits
- White House AI vetting: The Trump administration is weighing an executive order that would create a government review process for new AI models before public release, triggered largely by concerns over Anthropic’s Mythos and its ability to find thousands of software vulnerabilities. Kevin Hassett indicated a decision could come within two weeks, though the White House called reports “speculation.” The Hill
- EU AI Act streamlined: The EU Council and Parliament reached a provisional agreement to simplify high-risk AI system rules before the August 2, 2026 enforcement deadline, extending the sandbox establishment deadline to August 2027. EU Council
- U.S. state AI bill explosion: The U.S. now has over 1,200 active AI bills across state legislatures with no coherent testing framework, creating a compliance nightmare for companies operating nationally. Colorado’s SB 189 was just signed into law, replacing the original Colorado AI Act. Fortune
- Gemma open models: Google is using I/O day two (today) to detail new additions to its Gemma open-source model family, with a dedicated session on what’s new.
- Apple’s privacy AI push: Ahead of WWDC, reports indicate Apple’s revamped Siri for iOS 27 will include options to auto-delete chat histories after 30 days or one year, leaning into privacy as a differentiator for its AI features. Apple Insider
Worth Watching
The Mythos ripple effect continues. Anthropic’s model has now triggered a White House policy review, prompted the Bank of England to request a meeting with Anthropic’s leadership, and raised questions about whether a small number of Western firms having exclusive access to powerful vulnerability-scanning AI creates its own security risk. About 40 organizations have Mythos access — most central banks and governments do not. This story has legs well beyond the tech press.
Google’s agent play versus everyone else’s. Gemini Spark and the broader agentic push from I/O puts Google squarely in competition with OpenAI’s Codex agents and Anthropic’s managed agents. The difference: Google is leading with consumer-facing agents that work across its app ecosystem, while OpenAI and Anthropic are focused on developer and enterprise workflows. Worth watching which approach gains traction first.