AI Circle Daily
https://aicircle.news/archive/2026-08-19?lang=en
Scan to read this page
OpenAI paused training for its unreleased Astra model and overhauled internal safety protocols, tightening monitoring after its AI agents reportedly reached "critical" cyber capabilities.
Take: Talk about walking the talk on safety. This isn't just PR; it means their models actually hit a new, potentially dangerous capability threshold, forcing them to hit the brakes. The compute race is real, but so are the risks.
Anthropic's annualized revenue surged to $65 billion, adding $18 billion in just two months, indicating strong market momentum.
Take: This growth is insane, leaving OpenAI in the dust for now. Clearly, the market is hungry for Claude, or maybe OpenAI's recent drama created an opening for Anthropic to capitalize.
AI chip startup Etched doubled its valuation to $21 billion in under a month, after Jane Street, impressed by its first shipped AI cluster system, led a massive new funding round.
Take: $21 billion in a month, that's just printing money. When a top-tier quant firm like Jane Street not only uses your tech but also leads your round, you know Etched's AI chip solution is delivering real, measurable value.
Nvidia is investing $1.5 billion in SoftBank's data center developer, a move that guarantees its chips will power an OpenAI data center project.
Take: Smart play by Nvidia, locking in OpenAI's compute needs. This isn't just an investment; it's a strategic move to secure a major customer, guarantee market share, and further embed its chip ecosystem.
Amazon is reportedly destroying rare books to train its AI models, as these unique texts are highly valuable for LLM training beyond publicly available online data.
Take: The irony of Amazon, a company built on selling books, now destroying them for AI training is just chef's kiss. Can't license it? Destroy the original. It's a brutal move, but it highlights just how desperate companies are for high-quality, unique training data.
OpenAI introduced "ChatGPT for Teens," a dedicated mode with enhanced protections, healthy-use features, and parental controls, aiming to help teenagers learn and use AI responsibly.
Take: "For teens" is a nice spin, but let's be real, this is OpenAI reacting to mounting public scrutiny and regulatory pressure. It's a necessary step to mitigate risk and broaden their user base, albeit a bit late to the party.
China's powerful Z.ai AI model, previously warned about by experts, has been released, capable of both securing systems and potentially falling into the hands of hackers.
Take: The "dual-use" dilemma just landed. China's AI capabilities are clearly accelerating, and a model with both offensive and defensive potential like this is bound to send ripples through cybersecurity and even geopolitical strategy.
According to Hacker News, OpenAI has disbanded the team responsible for assessing the catastrophic risks of its models.
Take: So, they're beefing up safety *externally* but allegedly axing the "catastrophic risk" team *internally*? This move is perplexing. Either they think the big risks are gone, or they don't want anyone looking too closely. Contradictory much?
Groq raised $350 million at a $3.5 billion valuation, pivoting from an AI chipmaker to a "neocloud" business and expanding its Nvidia-powered data center footprint.
Take: Groq's pivot from hardcore AI chips to a "neocloud" business, leveraging Nvidia's hardware, is an interesting move. Maybe they realized the chip war is too brutal, and offering cloud services is a more viable path to sustained growth.
Cursor, known for its AI Code Editor, is launching a new code-hosting platform to capitalize on developer frustration and directly rival GitHub.
Take: GitHub is massive, but it's not without its critics. Cursor coming in with its AI-native DNA, deeply integrating AI with code hosting, could actually carve out a significant niche. It's a smart play against an entrenched incumbent.