AI News: June 25, 2026

1. Taktile Raises $110M to Automate High-Stakes Financial Decisions

Taktile. The agentic decision platform for financial institutions closed a $110M Series C led by Goldman Sachs, with participation from Tiger Global, Index Ventures, and Y Combinator. Taktile’s platform lets banks and insurers combine AI agents, business rules, and human oversight to automate decisions such as loan underwriting, fraud detection, and insurance claims, reporting outcomes including 95% automation in B2B underwriting and 75% fewer AML false positives. The round reflects how agentic AI is now considered mature enough for regulated, high-stakes financial workflows. Source

2. Assort Health Raises $120M at $1.2B Valuation for Healthcare Voice AI

Assort Health. The patient-access AI platform raised a $120M Series C led by Menlo Ventures, bringing total funding to over $222M and valuing the company at $1.2 billion. Its Synapse model was trained on 190 million patient voice interactions, 62,000 care protocols, and 1.6 million clinical decision pathways, enabling specialty-specific handling of scheduling, intake, eligibility, and medication refills. The company reported 20x revenue growth over 15 months and a 115% administrative labor capacity increase at partner health systems including John Muir Health. Source

3. Engram Raises $98M Seed to Cut Enterprise AI Token Costs

Engram. The eight-month-old AI memory startup emerged from stealth with a $98M seed round led by General Catalyst, at a $600M valuation, with just 13 employees. Engram’s technology aims to dramatically reduce the token costs enterprises pay when running AI at scale, and it has already secured early partnerships with Microsoft, Notion, and legal AI startup Harvey. The round is one of the largest seed rounds in AI infrastructure, underscoring investor conviction that token-cost reduction is a critical enterprise problem. Source

4. Agility Robotics Targets $2.5B Valuation in SPAC Deal

Agility Robotics. The humanoid robot maker announced plans to go public through a SPAC merger with Churchill Capital Corp XI at a $2.5B valuation, expecting to raise over $620M including $200M from institutional investors. Its bipedal robot Digit is deployed across nine customer sites with partners including Amazon, Schaeffler, and Toyota Motor Manufacturing Canada, and the company holds over $300M in multi-year orders. The listing reflects growing conviction from backers including NVIDIA and SoftBank Vision Fund 2 that humanoid robots are approaching commercial scale. Source

5. Data Shows Engineering Jobs Are Among the Most AI-Resilient Roles

SignalFire. New analysis from hiring intelligence firm SignalFire tracked millions of employees across 80 million companies and found engineering roles declined only 11% from 2019 levels, compared to a 25% drop in total tech hiring. Engineers comprised 55% of new hires at major tech companies in 2025, up from 46% in 2019, and early-stage startups hired 7% more engineers than before the AI wave. The findings are consistent with the Jevons paradox: AI automation expands the scope of available work rather than eliminating it, at least for now. Source

6. AI Researcher Exodus From Google Accelerates

Google. Two senior AI researchers who contributed to Gemini, Jonas Adler and Alexander Pritzel, are leaving Google for Anthropic, following recent departures including Noam Shazeer (who went to OpenAI) and Nobel Chemistry laureate John Jumper (who also went to Anthropic). The pattern suggests Anthropic and OpenAI are aggressively using pre-IPO equity to recruit Google’s top-tier talent. For practitioners, the concentration of talent at Anthropic and OpenAI could further widen the capability gap between frontier labs and everyone else. Source

7. Cerebras Shares Drop 20% After First Public Earnings

Cerebras. Despite reporting Q1 revenue of $193M (94% year-over-year growth) and a narrowing net loss, Cerebras shares fell nearly 20% after the company guided for full-year gross margins of 38-41%, well below the 47% achieved in Q1. CEO Andrew Feldman attributed the guidance to a plan to rent back equipment from a major customer while building out its own data center infrastructure, temporarily compressing margins. The stock drop, approaching IPO price, illustrates how investor confidence in AI hardware companies remains sensitive to margin trajectories even when revenue is strong. Source

8. Companies Reverse Course on AI Spending After Encouraging “Tokenmaxxing”

Industry. Companies that had promoted unfettered AI usage among employees, some building internal leaderboards to encourage adoption, are now imposing spending controls as costs prove unpredictable. Accenture is restricting employees from using AI for routine tasks like PDF conversion despite having recently tied promotions to AI adoption. The trend signals a broader industry shift into what analysts are calling the “era of token rationing,” in which AI platforms must demonstrate concrete ROI rather than relying on novelty to sustain enterprise budgets. Source

9. Europe Pushes Back on U.S. MATCH Act Targeting ASML Chip Sales to China

Policy. Dutch Trade Minister Sjoerd Sjoerdsma visited Washington to oppose the proposed MATCH Act, which would extend existing export controls to include ASML’s older deep ultraviolet immersion machines currently sold legally to China. ASML, the world’s sole manufacturer of advanced lithography equipment, derives 19% of its system sales from China, and the bill could significantly impact its revenue while straining transatlantic relations. For AI practitioners, disruption to the global semiconductor supply chain that starts with lithography tools could ripple into GPU supply and AI compute costs. Source

10. Micron Revenue Quadruples to $41B as AI Drives Memory Chip Shortage

Micron. The largest U.S. memory chip maker reported Q3 revenue of $41.45B, up roughly 4x year-over-year, with profit surging from $1.88B to $28.2B as AI model inference drives insatiable demand for high-bandwidth memory. Shares jumped 13% and the company’s market cap reached $1.2T. The shortage is expected to persist through 2027, and Micron recently struck a chip supply deal with Anthropic, positioning it as a critical partner to frontier AI labs. Source

11. GLM-5.2 Matches Claude Opus 4.7 on Coding Benchmarks at One-Sixth the Cost

Zhipu AI. Snowflake CEO Sridhar Ramaswamy published internal benchmark results showing that Chinese model GLM-5.2 achieved nearly identical three-attempt success rates to Claude Opus 4.7 on 103 programming tasks (66% vs. 67%), at a price of $1.40 per million input tokens versus Opus’s $5.00, and $4.40 versus $25.00 per million output tokens. Opus held a meaningful edge on first-attempt accuracy (53.7% vs. 47.6%) and token efficiency, but the cost differential creates significant pricing pressure on Western frontier labs in coding use cases. Source

Mistral. The company released OCR 4, a document recognition model supporting 170 languages that identifies the semantic role of each page element (titles, tables, equations, signatures) in addition to extracting text. In Mistral’s own testing across 600+ documents, independent reviewers preferred OCR 4 over competitors 72% of the time. The model is priced at $4 per 1,000 pages ($2 for batch), available via Mistral’s API, Mistral Studio, and Microsoft Foundry. Source

13. Pangram CEO: LLMs Give Themselves Away Through Argument Clustering

Pangram. Max Spero, CEO of AI text detector Pangram, explained that language models consistently generate arguments that cluster in a narrow conceptual band, whereas human arguments spread much more broadly across the possibility space. Even without interpretable rules, Pangram’s black-box classifier exploits this uniformity as a statistical fingerprint. The finding has direct implications for anyone building systems that need to distinguish human from AI-generated content, or for understanding where AI reasoning systematically differs from human reasoning. Source