The frontier model market repriced itself inside 72 hours. OpenAI shipped the GPT-5.6 family (Sol, Terra, Luna) on 9 July, xAI launched Grok 4.5 the same morning, Meta moved to paid distribution with Muse Spark 1.1, and a credible leak puts Gemini 3.5 Pro at general availability on 17 July. Against that backdrop, the EU AI Act reaches full applicability in 21 days, and the talent market has flipped: JPMorgan is now explicitly prioritising AI hires over bankers, and AI governance has become the fastest-growing hiring family in financial services.
- EU AI Act full applicability lands 2 August — transparency and high-risk obligations bite in three weeks; governance capacity, not model capability, is now the binding constraint
- Frontier price war is live — near-frontier inference floors dropped to $1–2 per million input tokens this week; Q1 build-vs-buy maths is already stale
- Banks' hiring flip is structural — 2,000+ AI roles across the six largest US banks in 12 months; governance and model-risk roles growing fastest
- JADEPUFFER sets a precedent — first documented end-to-end autonomous AI ransomware; expect security budget reallocation
The most consequential development of the past week is the simultaneous compression of frontier capability and price: three labs shipped or priced new frontier models within one news cycle, with Google expected to follow on 17 July. For enterprises — banks especially — this collides with the EU AI Act's 2 August deadline. The market is signalling the same conclusion through hiring: the fastest-growing AI roles are no longer builders but governors.
- OpenAI ships GPT-5.6 Sol / Terra / LunaHigh confidence
Launched 9 July across ChatGPT, API and Codex in three tiers: Sol at $5/$30 per million tokens (frontier), Terra at $2.50/$15 (production), Luna at $1/$6 (budget). 9 July was the first day on record with three frontier labs launching publicly accessible frontier models simultaneously.
- EU AI Act: T-minus 21 daysHigh confidence
Full applicability arrives 2 August 2026 — transparency duties and high-risk system rules included. The AI omnibus agreement extended high-risk rules for embedded products to 2028 and widened the SME simplified regime to firms up to 750 employees / €150M revenue.
- xAI launches Grok 4.5High confidence
Shipped 9 July, trained jointly with Cursor, priced at $2/$6 per million tokens and ranked fourth on Artificial Analysis's intelligence index.
- Gemini 3.5 Pro leak points to 17 July GAModerate confidence
Leaked launch plans indicate a 2-million-token context window, Deep Think reasoning on the $250/month Ultra tier, and API pricing near $1.25/$10. Unconfirmed by Google.
- JPMorgan prioritises AI hires over bankersHigh confidence
The bank's 2026 technology budget is $19.8B (up ~10%), with ~$1.2B toward AI-related investment. Nomura's AI strategy chief Tahir Zafar starts this month.
- JADEPUFFER: first autonomous AI ransomwareHigh confidence
Sysdig's Threat Research Team published the full analysis of the first documented end-to-end autonomous AI ransomware attack — a category-defining incident for enterprise security posture.
- Publishers seek sanctions against OpenAIHigh confidence
Newspapers led by The New York Times asked a federal court to sanction OpenAI for allegedly misrepresenting its ability to search models and output logs for copyrighted material.
- Illinois signs AI Safety Measures ActHigh confidence
Illinois joined the growing list of state AI regimes while the White House drafts voluntary AI release standards — the federal/state divergence raises multi-jurisdiction compliance cost.
Frontier labs & models
The week's defining feature is convergence: four labs shipping within days of each other, with pricing as the primary weapon. Luna at $1/$6 and Grok 4.5 at $2/$6 put near-frontier capability below the price of last year's mid-tier. Meta's move to paid distribution ends the era of free frontier-adjacent weights as a pricing anchor.
Enterprise adoption
Adoption pressure is shifting from experimentation to industrialisation: three-quarters of enterprises plan agentic deployment within two years, and the CAIO wave (43% of the Fortune 500, from 19% a year ago) shows accountability being formalised at board level. Financial services leads with roughly 62% CAIO density.
Policy & regulation
Three regulatory clocks are running: the EU AI Act's 2 August full applicability, the US state patchwork (Illinois latest), and White House voluntary release standards in draft. For multinationals the EU deadline dominates.
Security & risk
JADEPUFFER converts a theoretical risk into a documented incident class. Combined with the EU's cybersecurity-and-AI action plan of 7 July, expect security and AI-governance budgets to converge — and a hiring premium on people who can span both.
Banks & financial services
| Role | Who's hiring | Signal | Driver |
|---|---|---|---|
| AI/ML Engineer (front office) | Goldman Sachs (GBM), JPMorgan, Morgan Stanley | 2,000+ AI roles posted across the six largest US banks in 12 months; postings +143% YoY | Agentic workflow build-out in markets & banking |
| AI Governance / Responsible-AI Lead | Tier-1 banks broadly; 119 open GRC-AI roles tracked | Fastest-growing AI role family; anchored to NIST AI RMF, ISO/IEC 42001 | EU AI Act full applicability, 2 Aug |
| Model Risk Specialist (AI) | US & UK tier-1 banks | Averages ~$99–120k, top quartile above $160k and rising | Regulator expectations on model validation |
| Head of AI / Chief AI Officer | Citi (ex-IBM), CommBank (ex-Lloyds), JPMorgan (Zafar, ex-Nomura) | Executive imports from vendors and rivals; ~13% rise in AI headcount | Board-level AI accountability |
| Agentic / LLM Platform Engineer | JPMorgan (global AI rollout), Morgan Stanley, Barclays | ~$1.2B of JPMorgan's $19.8B tech budget directed at AI projects | Moving pilots into regulated production |
Enterprise (general)
The Chief AI Officer has gone from novelty to default: 43% of the Fortune 500 now have one (19% a year ago), 73% plan to by year-end, and recent appointments cluster in retail and health (Target, UnitedHealth, American Express, Lowe's; Deloitte created a global CAIO). Base compensation runs $280–650k. Below the C-suite, AI Engineer is LinkedIn's #1 fastest-growing US title; over 75% of AI-engineering postings now require domain specialisation.
| Item | Signal | Assessment |
|---|---|---|
| Together AI — $800M Series C | Post-money $8.3B | Inference-infrastructure consolidation; capacity pricing power ahead |
| Norm AI — $120M Series C | Khosla-led, 7 July | Compliance-AI category validated three weeks before the EU deadline — directly relevant to banks |
| Prime Intellect — $130M Series A | Radical-led, 8 July | Decentralised training attracting serious capital |
| H1 2026 global venture: $510B | Record half (Crunchbase) | AI-led boom accelerating funding and exits |
| API price floor | Luna $1/$6 · Grok $2/$6 · Muse $1.25/$4.25 | Near-frontier inference cost fell ~50% in a quarter — renegotiate vendor contracts now |
- Gemini 3.5 Pro on 17 July. Confirmation of the leak — 2M-token context near $1.25/$10 — would likely trigger matching price cuts from rivals within days.
- Bank earnings season opens ~14 July. Listen for AI headcount ratios and efficiency language on JPMorgan, Citi and Goldman calls.
- Commission guidance before 2 August. Documentation templates would ease the compliance sprint; continued silence raises the odds of a messy enforcement start.
- Sanctions ruling in publishers v. OpenAI. Sanctions would reset discovery and log-retention expectations for every model provider.
- A second JADEPUFFER. Any confirmed copycat autonomous-ransomware incident converts a one-off into a category and accelerates AI-security hiring immediately.