AI-found bugs drive a record vulnerability spike; Together AI raises $800 million

Research group Epoch AI reports that disclosures of high- and critical-severity software vulnerabilities from 21 major vendors — including Microsoft, Google, and Apple — hit roughly 1,500 in June 2026, more than three-and-a-half times the previous monthly record. Epoch links the surge to frontier models that can now autonomously find security flaws: Anthropic's Claude Mythos and OpenAI's Daybreak, whose partners had already flagged more than 10,000 critical bugs before public release. Epoch cautions the timing is correlation, not proof — part of the jump may reflect heightened interest in bug-hunting rather than raw capability alone.

Together AI, the cloud platform that hosts open-weight models like Llama, DeepSeek, and Mistral for developers, closed an $800 million Series C led by Aramco Ventures, with NVIDIA, Vista Equity Partners, and General Catalyst joining. The round values the company at $8.3 billion — a two-and-a-half-times step-up from early 2025 — and follows more than $1.15 billion in annual bookings. The raise is a bet that enterprises will keep shifting inference workloads onto cheaper open models, and Together plans to grow its compute footprint roughly fifty-fold over the next five years.

Google DeepMind and the independent film studio A24 announced a research partnership that pairs the lab's researchers with A24's filmmakers to develop new creative tools and production workflows, with Google also taking an undisclosed equity stake in the studio. The companies named no specific models and set no fixed deliverables, framing it as open-ended, exploratory work. It is a notable move as frontier labs court Hollywood, arriving amid ongoing tension over AI's role in creative work.

Elon Musk imposed a $200-per-week cap on Tesla employees' AI-tool spending, requiring manager sign-off to go over, according to The Information. Notably, the limit exempts his own Grok as well as Anysphere's Cursor, which Musk has pushed staff to use — a pointed carve-out for favored tools. Tesla joins Uber, Meta, Walmart, and Coinbase in reining in runaway AI bills, the latest sign that the "use all the AI you can" era is giving way to hard cost discipline.

Mistral released Leanstral 1.5, an Apache-2.0 open model for formal mathematical proof in Lean 4. The mixture-of-experts design carries 119 billion total parameters but activates just 6 billion, and it scores a perfect 100 percent on the miniF2F benchmark while solving 587 of 672 problems on PutnamBench. Beyond math, the model flagged five previously unknown bugs across 57 open-source repositories during testing — including a critical integer-overflow flaw in a decoding library — pointing at formal verification as a practical path to safer software.

Portugal launched Amália, its first government-backed open-source large language model, purpose-built for European Portuguese and released with its weights, datasets, and source code. Built by a consortium of Portuguese universities on top of the European EuroLLM foundation model and funded with 5.5 million euros, the 9-billion-parameter Amália adds text and image understanding. Named after the fado singer Amália Rodrigues, it is part of a broader European push for sovereign AI that reflects local language and culture rather than depending on foreign systems.