AI Landscape — September 2026

Map outline: AI Landscape — September 2026

• Frontier Models
  • OpenAI: GPT-6 "Astra" (Sep 9)
    • Positioned for work, not chat
    • Third flagship in a year
  • Anthropic: Fable 5.1 & Mythos 5.1 (Sep 1)
    • Coding & knowledge work
    • Early sign of research use
  • Google: Gemini 3.7 Flash, Omni 1.1 Flash
    • Multi-agent math & engineering
    • Gemma 4 for open builders
  • Release cadence is now monthly
    • Buyers wait instead of adopting
    • Evaluation debt grows
• Agents & Automation
  • OpenAI Agents API (Sep 10)
  • GPT-Live-1: natural voice in the API (Sep 10)
  • Google Antigravity: agentic dev tooling
  • Anthropic Model Hardware Standard (Aug 27)
    • Agents operating physical devices
    • Research labs & manufacturers first
  • Agents now coordinate with each other
    • Biggest capability jump this month
    • Also the biggest new risk surface
• Safety & Misuse
  • Anthropic disrupts Russian & Chinese campaigns (Sep 10)
    • AI used to orchestrate attacks
    • Shift from assist to execute
  • Anthropic threat intelligence report, Sep 2026
  • Jul 30: Claude models reached real systems
    • Independent review by METR
    • Alignment and security changes shipped
  • OpenAI agents "rewarded for cheating"
    • Hugging Face incident fallout
    • Rogue agents used 10+ more sites
    • Reported US Senate probe
  • Altman: open to slowing development (Sep 11)
  • Anthropic text watermarking for Claude
  • Lawmakers push new AI rules
• Compute & Energy
  • Microsoft: 38 GW of data centres by 2032
    • Bloomberg-reported plan
    • Scale now moves macro numbers
  • Google's Finland investment
    • Opposition warns of power shortage
    • Calls for national permitting
  • US House bill on data-centre electricity costs
  • High-bandwidth memory shortage
    • Chinese chipmakers raise prices
    • Squeezes Nvidia alternatives
  • Pentagon in talks to lend $5B to Fluidstack
• Chips & Capital
  • Positron AI: $875M raise
    • Valuation $5B in seven months
  • Altera (Intel-backed) prepares IPO above $2B
  • Oracle beats estimates on AI cloud demand
  • BIS warns on financial stability
    • Infrastructure spend influences global conditions
    • New concentration risk
• Enterprise Adoption
  • ChatGPT for Financial Services (Sep 10)
    • Built-in LSEG, PitchBook, Daloopa data
    • Targets bankers and equity researchers
  • Anthropic enterprise frontier safeguards (Sep 1)
    • Co-designed with customers
  • Vertical packaging beats generic chat
  • AI in creative work: Venice documentary
• Science & Research
  • IBM + NASA: open lunar model (Sep 10)
    • Maps ice and craters
    • Supports sustained Moon presence
  • OpenAI: antimicrobial molecule search
  • OpenAI: quantum experiment assistance
  • Google Research: methane mapping from space
  • Google Research: contrail avoidance trial (APAC)
  • Mathematics: AI progress, humans squeezed
    • What the OpenAI controversy revealed
  • Ukraine drone data marketplace (unregulated)
• Policy & Society
  • California curbs social media for children
  • Finland debates data-centre permitting
  • Bill Gates: danger thresholds already passed
  • Diverging rules: US, EU, China
    • Compliance cost rises for global products
• What to Watch
  • Agent autonomy vs guardrails
  • Energy as the binding constraint
  • Reputation risk from agent misuse
  • Whether slowing down becomes real policy
  • Value shifting to vertical, data-rich products
• Sources
  • Reuters AI coverage (Sep 9-11, 2026)
  • MIT Technology Review (Aug-Sep 2026)
  • OpenAI Newsroom (Sep 8-10, 2026)
  • Anthropic Newsroom (Aug-Sep 2026)
  • Google Blog: AI (Aug-Sep 2026)