Cerebras Launches CS-4: A 750-PetaFLOP Monster That Challenges Nvidia's GPU Dominance
Cerebras Systems dropped a bombshell at its Supernova event in San Francisco on August 19, unveiling the CS-4 — a rack-scale AI accelerator that packs three WSE-3 Turbo wafer-scale processors into a single system. The result: 750 petaFLOPS of AI compute and 129.6 petabytes per second of memory bandwidth.
The company claims the CS-4 delivers up to 30 times faster inference than competing GPU-based systems, positioning it as a direct challenge to Nvidia's dominance in the AI infrastructure market. Compared to its predecessor, the CS-3, the new system offers twice the speed and 10x more throughput per watt — a critical metric as data center energy consumption comes under increasing scrutiny.
A new modular "backpack" architecture, branded Nexus, reduces component count by 50 percent and simplifies deployment. First shipments are expected later this quarter, and early interest from hyperscalers suggests Cerebras is no longer content to be a niche player in the AI chip race.
Sources: Cerebras Blog, Bloomberg, HPCwire
Jeff Dean Leaves Google After 27 Years to Co-Found Discovery Loop
In one of the most significant talent moves in AI history, Jeff Dean — Google's legendary Chief Scientist — announced his departure on August 5 to co-found Discovery Loop, an AI-for-science startup. He's not leaving alone: Sanjay Ghemawat (Google Fellow, 26 years), Oriol Vinyals (VP of Research at DeepMind), and Quoc Le (Google Brain co-founder) are all joining the venture.
Discovery Loop is structured as a public benefit corporation focused on using AI to automate scientific and engineering research — starting with machine-learning research itself before expanding into drug discovery, hardware design, and clean energy. Radical Ventures and Khosla Ventures are co-leading the seed round, with Google staying involved as a founding investor and cloud partner.
The departures are part of a broader Google DeepMind restructuring. Demis Hassabis is stepping down as DeepMind CEO to become chairman and Alphabet's chief scientist, while CTO Koray Kavukcuoglu takes over Gemini model development as a newly elevated SVP. The reshuffle signals that even the most talent-rich AI lab isn't immune to the gravitational pull of startup ambitions.
Sources: TechCrunch, Unite.AI, GeekWire
"CoSnitch": Microsoft Patches Critical Copilot Vulnerability That Could Steal Your Data With One Click
Security researchers at Varonis Threat Labs disclosed CVE-2026-24301, a critical vulnerability in Microsoft Copilot Personal dubbed "CoSnitch" that could silently exfiltrate data from connected Gmail, Google Drive, and Calendar accounts — all triggered by a single click on a malicious link.
The attack chain exploited three linked flaws: an undocumented prompt autorun mechanism that executed attacker-supplied prompts in a victim's authenticated Copilot session, abuse of Copilot's existing permissions to pull sensitive data, and a separate persistent memory-poisoning path through web summarization. Essentially, it turned Copilot's broad access to connected services into an attack surface.
Perhaps most striking was the discovery method: researchers used "meta-hacking" — repeatedly questioning Copilot about why automatic execution wasn't possible, reframing each refusal as a follow-up question, effectively social-engineering the AI's reasoning process rather than attacking code directly. Microsoft patched the flaw on August 18 and found no evidence of exploitation in the wild.
Sources: The Hacker News, Cybersecurity News, NVD
EU AI Act Transparency Rules Take Effect — With Real Penalties
As of August 2, the European Commission's AI Office and national authorities began actively enforcing the EU AI Act's transparency provisions. The rules require AI systems to clearly disclose when users are interacting with AI rather than a human, and mandate that deepfakes and AI-generated content be labeled as such.
The penalties are designed to bite: noncompliance can trigger fines of up to €15 million or 3% of worldwide annual turnover, whichever is higher. While enforcement has only just begun, the rules represent the most concrete regulatory action on AI transparency anywhere in the world.
Meanwhile, the higher-stakes provisions covering "high-risk" AI systems have been pushed back — stand-alone high-risk systems won't face compliance deadlines until December 2027, and those embedded in products get until August 2028. In a parallel development, China became the first country to regulate AI agents as a distinct category, with its Implementation Opinions becoming enforceable on July 15.
Sources: European Commission, Cooley LLP, Eve AI Core
OpenAI's IPO Inches Closer as Anthropic Hits Profitability Two Years Early
The AI industry's two leading foundation-model companies are both hitting financial milestones that would have seemed impossible two years ago. OpenAI's confidential S-1 filing, submitted to the SEC on June 8, is expected to become public in late August, with a September listing target at a valuation between $850 billion and $1 trillion. The company is generating roughly $25 billion in annualized revenue with over 230 million weekly ChatGPT users — though it still loses approximately $1.22 for every dollar earned.
On the other side, Anthropic reported Q2 2026 revenue of $10.9 billion (up 130% year-over-year) and its first-ever operating profit of $559 million — arriving two years ahead of its own internal projections. The achievement marks a rare milestone: profitability in the capital-intensive foundation model business.
Together, the two stories paint a picture of an industry that is rapidly maturing financially, even as questions about long-term unit economics and the sustainability of current spending levels remain very much open.
Sources: Fortune, AI Tools Recap, Investing.com