Anthropic projects $10.9B in Q2 revenue and ~$559M operating profit, becoming the first major AI lab to reach profitability — roughly 2 years ahead of its own internal forecast. This validates that frontier model companies can build sustainable unit economics, not just burn VC capital. It also signals API pricing and enterprise adoption have reached scale.
A coalition of researchers is suing the Trump administration to protect their ability to study hate speech, disinformation, and online harassment — research that has been directly targeted by government pressure since early 2025. The lawsuit's outcome could define what third-party AI safety and content research is legally permissible in the US. Global implications are significant given US platform dominance.
Duplicate coverage of Anthropic's projected $10.9B Q2 revenue and $559M operating profit, citing the Wall Street Journal as the original source. Notably adds that Anthropic didn't expect profitability until 2028 as recently as last summer, suggesting enterprise API adoption has dramatically outpaced internal projections. The main drivers are API revenue from enterprise and developer customers.
TeamPCP is executing supply chain attacks on open source repositories at an unprecedented scale, with GitHub as the latest vector. This is not a novel threat class, but the scale and coordination represent a meaningful escalation that affects any team pulling open source dependencies — including AI/ML tooling. The blast radius for AI builders is unusually high given the heavy reliance on PyPI, Hugging Face, and GitHub-hosted model weights.
The US government is taking direct equity positions totaling $2B across nine quantum computing startups, marking a shift from grants to ownership-based public investment in deep tech. One beneficiary is backed by a firm with reported Trump family connections, raising governance questions. For AI builders, quantum is still 5-10 years from practical utility, but government capital concentration will shape which players survive to productization.
An OpenAI reasoning model disproved the Erdős unit-distance conjecture — open since 1946 — using algebraic number theory techniques that experts hadn't anticipated, producing a proof accepted by a top mathematics journal. Fields Medalist Tim Gowers has publicly validated the result. This is the first AI-generated proof to clear peer review at mathematics' highest tier, marking a qualitative shift in what automated reasoning can produce.
Sebastian Raschka's deep-dive covers KV cache sharing, multi-head compression (mHC), and compressed attention mechanisms appearing in recent open-weight models including Gemma 4 and DeepSeek V4. These architectural innovations directly reduce inference memory and compute costs for long-context workloads. This is the highest-engagement technical piece in this set (HN: 37) and represents the state-of-the-art in production-oriented architecture research.
OpenAI's reasoning model disproved the 80-year-old Erdős unit-distance conjecture in discrete geometry, producing a peer-review-quality mathematical proof accepted by a top journal — the first of its kind. With an HN score of 2417, this is the breakout story of the week by engagement margin. The result demonstrates that frontier reasoning models can now generate novel, verifiable knowledge rather than recombining existing knowledge.
That's today's briefing.
Get it in your inbox every morning — free.
Help us improve AI in News
Got a suggestion, bug report, or question?