# AI News Flash — Daily Brief **Date:** Wednesday, June 10, 2026 **Type:** daily **Source:** https://ainewsflash.co/brief/29 **Editors:** Justin Bunnell (https://www.linkedin.com/in/justinbunnell/), Laz Manrique (https://www.linkedin.com/in/laz-m-5a218b81/) **Publisher:** AI News Flash (https://ainewsflash.co) **License:** Republish with attribution. --- ## Key takeaways - Anthropic launches Claude Fable 5, a new tier above Opus with built-in safety routing. - OpenAI runs its first live ad test inside ChatGPT for UK free and Go users. - Apple AFM 3 Core Advanced runs a 20B sparse multimodal model entirely on iPhone. - Google Gemma 4 12B handles audio, video, and images without separate encoders. - Senate votes 99-1 to remove a 10-year ban on state AI regulations from federal law. - Colorado replaces its AI Act with a narrower law taking effect January 1, 2027. --- ## Platforms ### Anthropic launches Claude Fable 5, a new tier above Opus with built-in safety routing. Anthropic released Claude Fable 5 on June 9, introducing a new model tier above Opus and priced at $10 per million input tokens and $50 per million output tokens, exactly double the Opus 4.8 rate. The model ships with hard safety classifiers that automatically reroute sensitive cybersecurity and bioweapons queries to Opus 4.8 rather than processing them directly. A parallel variant called Claude Mythos 5, built on the same underlying weights but with those safeguards removed, remains locked to Project Glasswing partners and is not publicly available. All paid Claude plan subscribers receive free access to Fable 5 through June 22. > Why it matters: Enterprises evaluating frontier models must now account for built-in query routing that limits direct access to certain sensitive use cases. - Source: https://letsdatascience.com/news/anthropic-releases-claude-fable-5-for-broad-use-db880a41 ### OpenAI runs its first live ad test inside ChatGPT for UK free and Go users. OpenAI has started rolling out advertisements inside ChatGPT for users on its Free and Go plans in the United Kingdom, marking the company's first live deployment of ad-supported monetization. Subscribers on Plus, Pro, Business, Enterprise, and Education plans are not affected and will continue to receive an ad-free experience. The move signals that OpenAI is actively exploring revenue streams beyond subscriptions and API fees, a strategic consideration that analysts have tied to the company's anticipated IPO. The UK rollout functions as a controlled test before any broader expansion. > Why it matters: AI developers and enterprises should monitor how ad integration affects user experience benchmarks and OpenAI's competitive positioning against ad-free alternatives. - Source: https://releasebot.io/updates/openai/chatgpt ## Capabilities ### Apple AFM 3 Core Advanced runs a 20B sparse multimodal model entirely on iPhone. At WWDC 2026 on June 8, Apple unveiled AFM 3 Core Advanced, a 20-billion-parameter sparse model that activates only 1 to 4 billion parameters per request by storing weights in flash memory and loading only the relevant parameters at inference time. It runs entirely on-device on iPhones with 12 GB RAM, a significant jump beyond the prior 3-billion-parameter on-device ceiling. The model handles text and images natively without routing data to the cloud. Apple also expanded the framework beyond iOS, adding Linux support and a Python SDK that opens the toolchain to a much wider developer audience. > Why it matters: On-device sparse inference at this scale gives developers a privacy-preserving, cloud-free path for building multimodal applications on consumer hardware. - Source: https://www.macstories.net/linked/the-third-generation-of-apples-foundation-models-and-afm-core-advanced/ ## Technology & Research ### Google Gemma 4 12B handles audio, video, and images without separate encoders. Released on June 3, Google's Gemma 4 12B is an encoder-free 12-billion-parameter open model that projects raw audio waveforms and visual patches directly into the LLM embedding space through lightweight linear layers, removing the separate vision and audio encoders that typically increase latency and memory usage. The model fits within 16 GB of RAM, ships with a 256K context window, and includes speculative-decoding multi-token-prediction drafters to accelerate inference. It handles text, images, video, and audio in a single forward pass, making it the first mid-sized open Gemma with native audio support and a practical foundation for fully local multimodal agents. > Why it matters: Developers building local multimodal agents now have an open, laptop-friendly model that handles audio natively without the overhead of separate encoder components. - Source: https://blog.google/innovation-and-ai/technology/developers-tools/introducing-gemma-4-12b/ ## Regulation & Policy ### Senate votes 99-1 to remove a 10-year ban on state AI regulations from federal law. The Senate voted 99-1 to remove a provision from the One Big Beautiful Bill Act that would have preempted state AI regulations for ten years, with Senator Thom Tillis of North Carolina casting the sole dissenting vote. The bill was signed into law without any restriction on state-level AI legislation, delivering a direct rebuke to the White House's push for federal preemption. It marks the second time Congress has rejected a broad state-preemption approach to AI governance. With that avenue closed, the DOJ AI Litigation Task Force now stands as the administration's primary tool for challenging state AI laws. > Why it matters: Enterprises and AI developers must continue planning for a fragmented landscape of state-by-state AI regulations rather than a single federal standard. - Source: https://www.congress.gov/bill/119th-congress/house-bill/5388 ### Colorado replaces its AI Act with a narrower law taking effect January 1, 2027. Colorado's SB 26-189, signed on May 14, repeals the state's original AI Act and replaces its broad high-risk AI system framework with a narrower automated decision-making technology standard that takes effect January 1, 2027, granting affected businesses approximately seven additional months to prepare for compliance. The DOJ's intervention supporting xAI's constitutional challenge to the original law served as a direct catalyst for the legislature's accelerated pivot. Despite the law's lighter overall regulatory footprint, compliance teams should note that its automated decision-making technology definition covers a wider set of decision contexts than many organizations have recognized in their initial assessments. > Why it matters: Compliance teams in Colorado must update their AI governance frameworks to reflect the new automated decision-making standard before January 1, 2027. - Source: https://www.lawfuel.com/colorado-ai-act-2026-sb-26-189/ ## AI Stocks ### AMD stock doubles in 2026 as data center revenue hits a record $5.8B in Q1. AMD has more than doubled in stock price year-to-date in 2026, outperforming Nvidia, Palantir, and Alphabet as its Q1 data center revenue reached a record $5.8 billion, up 57% year-over-year, and total revenue rose 38% to $10.3 billion. CEO Lisa Su doubled the company's CPU market-growth forecast to more than 35% annualized through 2030, attributing the outlook to structural demand from agentic AI inference workloads. Custom MI450 GPUs built for Meta and OpenAI are scheduled to ship in the second half of 2026. Analysts now project 51% annualized EPS growth, with every major hyperscaler expanding EPYC CPU deployments alongside their GPU infrastructure. > Why it matters: AMD's surging data center revenue and custom GPU deals with Meta and OpenAI signal a meaningful shift in enterprise AI infrastructure spend away from single-vendor GPU dependency. - Source: https://www.fool.com/investing/2026/05/13/meet-the-ai-stock-crushing-palantir-nvidia-and-alp/ --- Archive: https://ainewsflash.co/archive RSS: https://ainewsflash.co/rss.xml LLM index: https://ainewsflash.co/llms.txt