Twenty-plus AI stories dropped between May 10 and 17. Most of them do not change what your team ships this week. Relve, a trusted AI intelligence platform, filtered through all of it so founders and their teams know what actually matters. Here is the full noise briefing - and what to keep on your radar.
Key Takeaways
- OpenAI ran three threads at once, Codex went mobile, Apple legal tensions surfaced, and Altman took the stand in the Musk trial
- Anthropic passed OpenAI in paid business adoption for the first time, per the Ramp AI Index across 50,000-plus companies
- Google replaced Chromebook with Googlebooks, an Android-based AI-native laptop line with Gemini built in at the foundation
- YouTube expanded deepfake detection to every adult on the platform, up from Partner Program creators and public figures only
- Hackers used AI to build and deploy a confirmed zero-day exploit – the first documented case, and the ones to watch this week reflect what comes next
Codex is now in the ChatGPT mobile app on iOS and Android, letting developers monitor tasks, review outputs, approve commands, and start new threads remotely. A new Chrome extension connects Codex directly into a live browser session, giving it access to signed-in apps, multiple tabs, and authenticated workflows.
Anthropic launched a comparable Remote Control feature for Claude Code in February. The browser session is now the key battleground for coding agents.
OpenAI has engaged an outside law firm to explore legal action against Apple over the ChatGPT integration announced at WWDC 2024. The partnership underperformed significantly – the integration was buried, features were hard to find, and subscriber revenue fell well short of projections.
Apple has since struck a separate deal paying roughly $1 billion a year for Google’s Gemini models. No formal notice yet, and any move is expected to wait until after the Musk trial concludes.
Sam Altman took the stand in the Musk vs OpenAI trial. He testified that Musk’s demand for personal control – including a suggestion the company pass to his children – was the central reason he was excluded from the for-profit restructuring. OpenAI’s nonprofit now holds equity valued at over $150 billion. The trial continues with no ruling. Noise until there is a verdict.
34.4% of Ramp-tracked businesses now pay for Anthropic, ahead of OpenAI at 32.3%, per the May Ramp AI Index across 50,000-plus companies. Anthropic’s paid adoption rose 26 percentage points in 12 months, from 9% in May 2025.
The index captures only paid transactions through Ramp’s platform – free tiers and personal accounts are not counted. Ramp’s own economist publicly flagged the lead may not hold, citing three underrated risks he did not name in the interview.
Googlebooks is Google’s Android-based AI-native laptop line, launching fall 2026 with Acer, Asus, Dell, HP, and Lenovo. Gemini is built in at the foundation, with a Magic Pointer cursor that surfaces contextual suggestions as users hover over screen elements.
On Android, Gemini now handles multistep tasks across apps, auto-browse, form filling via Personal Intelligence, and AI dictation via Rambler in Gboard. Create My Widget lets users describe custom home screen widgets in plain language. Enterprise and education buyers are the test – they are the core Chromebook constituency and the ones who will decide whether this transition lands.
YouTube opened its AI deepfake detection feature to every adult on the platform, up from Partner Program creators, politicians, journalists, and entertainment figures.
Enrollment requires a selfie scan and ID verification. Removal is not guaranteed and covers facial likeness only – voice clones fall outside its scope. Detection happens after content is already live. There is no pre-upload screening.
Markifact launched a Meta Ads MCP server letting marketers manage Facebook and Instagram campaigns from Claude, ChatGPT, Cursor, and other MCP-compatible clients. Approval-based controls mean changes are staged for human review before going live. Independent validation of its safety claims is not yet available. Treat it as a drafting layer, not an autonomous execution engine.
Osaurus is a Mac-only open-source LLM server with 112,000 downloads, connecting local and cloud AI models through a single interface. Hardware bar is steep – 64 GB RAM minimum, 128 GB recommended.
Wispr Flow says India is now its fastest-growing market, with 100% month-over-month growth after a Hinglish rollout – but India accounts for 14% of installs and only 2% of in-app purchase revenue.
Digg relaunched as an AI news aggregator ranking stories by X engagement in real time, in beta, described as intentionally raw and buggy.
Origin Lab raised $8 million to build a marketplace connecting video game companies with world-model AI labs – seed stage, no disclosed studio partners, no confirmed lab deals.
Why we’re watching these. These stories did not produce a ruling, a product change, or an immediate action item. But the patterns they are setting will shape how AI gets deployed, defended, and bought across enterprise teams this year.
Google’s Threat Intelligence Group confirmed the first known case of AI being used to discover and weaponize a zero-day vulnerability – a Python-based two-factor authentication bypass in a widely used open-source admin tool.
The exploit targeted a semantic logic flaw, the class of error traditional fuzzers and static analysis tools are designed to miss. State-linked actors from China and North Korea are running thousands of recursive prompts against CVE databases. The LiteLLM AI gateway was separately compromised in March. Organizations relying solely on traditional static analysis tools now have a documented blind spot.
Clio reached $500 million ARR, doubling from $200 million in under 18 months. The milestone lands the same week Anthropic expanded Claude for Legal, directly competing with Harvey ($190M ARR) and Legora ($100M ARR) – both of which rely on Claude as a core model.
Legal is compressing at the same rate code did in 2023. The supplier competitor dynamic is now impossible to ignore and neither firm has publicly addressed how they plan to navigate it.
Notion launched a developer platform turning its workspace into an agent orchestration layer – Notion Workers for custom cloud code, database sync from Salesforce, Zendesk, and Postgres, and an External Agent API supporting Claude Code, Cursor, Codex, and Decagon at launch.
Workers are free through August 2026. The risk is complexity: Notion’s appeal was always accessibility, and CLIs and sandboxed code execution bifurcate the audience.
Dessn raised $6 million to let designers iterate on live production codebases without local setup, positioning against Figma and ideation tools like Lovable and v0.
Clawdmeter hit 800 GitHub stars in four days – a hobbyist hardware token tracker for Claude Code that signals how deeply the tool has embedded itself in developer workflows.
Adaption’s AutoScientist claims to automate the full AI model training research loop, raising win rates from 48% to 64% over human-configured fine-tuning in its own internal benchmarks. No independent validation yet – the 30-day free window is an implicit acknowledgment the product needs to prove itself in real environments.
Missed last week’s roundup? Catch up here.
Those were last week’s noise stories. The announcements that actually change how AI is being deployed inside enterprise teams are a different category. Last week Relve published a signal most founders missed: Anthropic’s Mythos Changed the Security Baseline – the first production-scale proof that the false-positive problem blocking AI security tools has been solved. Browse all signals at Relve.
