Fact-checking site Snopes debunked an AI generated image of Kentucky Sen. Mitch McConnell shown covered in tubes in a hospital bed on Wednesday, noting the image registered as containing a SynthID watermark and had been produced by an AI system participating in Google's program, per TechCrunch. SynthID, launched at Google I/O in 2025, embeds an invisible signature in AI generated images that Gemini models have carried since launch; OpenAI joined the program in May 2026. The signature is built into the pixels themselves, so it survives when an image is screen captured across platforms as the McConnell image was. McConnell's actual health has been the subject of speculation since he checked into the hospital after an emergency call on June 14.
Read at TechCrunch ↗
SpaceXAI launched Grok 4.5 on Wednesday, its first model release since going public and acquiring Cursor, pitching it as a coding and agenticAI agentAn AI system that carries out multi-step tasks on its own, such as browsing, writing code or making purchases, rather than answering a single prompt. Agents raise new questions about liability, security and oversight because they act rather than just advise. work tool trained alongside Cursor on tens of thousands of Nvidia GB300 GPUs. Musk called it "an Opus-class model, but faster, more token-efficient and lower cost," and SpaceXAI priced it at $2 per million input tokens and $6 per million output tokens, against $5 and $25 for Anthropic's Opus 4.8. The company's own charts show mixed results: Grok 4.5 beats Opus 4.8 on two of four published coding benchmarks, DeepSWE 1.0 and Terminal Bench 2.1, and trails on DeepSWE 1.1 and SWE Bench Pro, while Anthropic's Fable leads across the board. Axios noted the model was trained on the same computeComputeThe processing power used to train and run AI models, usually measured in chips, GPU-hours or FLOPs. capacity SpaceXAI leases to Anthropic and Google. The model is unavailable in the EU until mid-July.
Read at Axios ↗ • Read at SpaceXAI ↗
RAND's Center on AI, Security, and Technology released a 20 page primer Thursday on High Bandwidth Memory, the specialized DRAMDRAMThe working memory that phones, PCs, and AI servers use to hold data that a processor is actively handling. used in most leading AI chips, finding that only SK Hynix, Samsung and Micron produce it at globally competitive scale and that South Korea will hold 78% of global HBM wafer capacity by the end of 2026. RAND estimates HBM can account for 45% of an AI chip's production cost, roughly three times the cost of fabricating its processor, and note the United States produces no HBM domestically, leaving it dependent on South Korea, Japan and Taiwan until at least 2030 despite SK Hynix's Indiana packaging plant (2028) and Micron's New York fab (2030). China's CXMT remains roughly four years behind the state of the art, making its progress the key variable for whether U.S. export controls keep Chinese AI chipmakers memory-bottlenecked.
Read at RAND ↗