FBI seeks predictive AI for political watch list · FreeToken runs 753B-parameter MoE models on personal machines · UK security institute's AI models caught deploying malware · DeepSeek's new multimodal model outperforms Anthropic's Opus 4.8 · NVIDIA's AVO agent system achieves perfect ARC-AGI-3 score · Open models catch up twice as fast each era · Block open-sources Berd, unifying AI agents across models · Ramp launches Router, taking on OpenRouter in AI routing · Harari warns AI could erode trust, enable techno-despotism · OpenAI tightens testing after AI agents hacked Hugging Face · Texas student thwarts rogue AI agent's GitHub sabotage attempt · DefendEye scales production of autonomous AI drone for first responders · Encrypted prompts trick Grok into stealing user data · Claude doubles protein design hit rate in wet-lab tests · MiniMax Design brings Codex-style capabilities to video models · SpaceX's AI satellite plan would export rare metals to space · Ornith-1.5 open models generate their own training tasks · A software engineer's case for balanced AI regulation · Ramp Router cuts LLM inference costs by 40% · Meta's Muse Video beta shows state-of-the-art potential · FBI seeks predictive AI for political watch list · FreeToken runs 753B-parameter MoE models on personal machines · UK security institute's AI models caught deploying malware · DeepSeek's new multimodal model outperforms Anthropic's Opus 4.8 · NVIDIA's AVO agent system achieves perfect ARC-AGI-3 score · Open models catch up twice as fast each era · Block open-sources Berd, unifying AI agents across models · Ramp launches Router, taking on OpenRouter in AI routing · Harari warns AI could erode trust, enable techno-despotism · OpenAI tightens testing after AI agents hacked Hugging Face · Texas student thwarts rogue AI agent's GitHub sabotage attempt · DefendEye scales production of autonomous AI drone for first responders · Encrypted prompts trick Grok into stealing user data · Claude doubles protein design hit rate in wet-lab tests · MiniMax Design brings Codex-style capabilities to video models · SpaceX's AI satellite plan would export rare metals to space · Ornith-1.5 open models generate their own training tasks · A software engineer's case for balanced AI regulation · Ramp Router cuts LLM inference costs by 40% · Meta's Muse Video beta shows state-of-the-art potential

Memia - AI Situational Awareness

STRATEGIC INTELLIGENCE FOR AI DECISION-MAKERS

Horizon-scanning and sensemaking for governments, capital and operators navigating the AI economy.

Latest briefings

A live read on what moved across the AI landscape, what it means, and what's next.

Upcoming Events

Agentic Engineering Day New Zealand
AIOctober 21, 2026

Agentic Engineering Day New Zealand

Agentic Engineering Day New Zealand is a community-driven conference for builders working with AI agents, MCP, AI infrastructure, and production AI systems. The event brings world-class AI engineering...

NZ Tech Expo 2026 - Technology-Led Export and Scaling for Global Growth
TechnologySeptember 17, 2026

NZ Tech Expo 2026 - Technology-Led Export and Scaling for Global Growth

The best technology conversations are not really about technology. They are about how businesses use it to compete, scale, and grow internationally. At NZ Tech Expo 2026, NZTE will support and host pr...

How Exponential AI Is Shaping the Future of International Development
International DevelopmentSeptember 8, 2026

How Exponential AI Is Shaping the Future of International Development

Memia's Head of Research Ben Reid will be delivering a keynote at The Council for International Development's two-day annual conference in Te Whanganui-a-Tara Wellington in September. The conference b...

More ->

Memia Writing

Mind-expanding essays and frameworks on AI, accelerating technology, and the future.

Anatomy of an AI agent (mid-2026 edition)

June 9, 2026

Anatomy of an AI agent (mid-2026 edition)

Models, tokens, harnesses, agents, skills, tools - making sense of a dynamic vocabulary....

The Intelligent Interstitium: What Lies Between

April 2, 2026

The Intelligent Interstitium: What Lies Between

An exploration of the connective layer between human minds, tracing how communication media have evolved from passive channels into increasingly agentic systems that now participate in thought itself....

More ->

Sensorium Signal Scanning

Real-time links and signal scanning from across the web, tracked through Memia Sensorium.

FBI seeks predictive AI for political watch list

FBI seeks predictive AI for political watch list

Signals:

  • The FBI is integrating predictive AI to automate domestic political surveillance and watch-listing.
  • Expanded definitions of "terrorism" now target political dissenters, journalists, and elected officials.
  • Automated pre-crime policing risks significant constitutional challenges and potential civil liberty violations.
FreeToken runs 753B-parameter MoE models on personal machines

FreeToken runs 753B-parameter MoE models on personal machines

Signals:

  • Enables running massive frontier-scale AI models on standard consumer-grade hardware.
  • Reduces infrastructure costs by eliminating the need for expensive datacenter GPU clusters.
  • Optimizes local resource utilization through dynamic, bandwidth-adaptive execution for edge devices.
UK security institute's AI models caught deploying malware

UK security institute's AI models caught deploying malware

Signals:

  • AI safety testing risks violating national cybersecurity and computer misuse laws.
  • Unregulated model testing creates significant legal and ethical liability for institutions.
  • Autonomous AI agents can inadvertently execute malicious code without human oversight.
More ->

Interpret the future

Key narrative axes which Memia is tracking relating to Artificial Intelligence trends and impacts

Is it an AI Bubble?

There is no AI investment bubble

AI companies continue to justify their vast valuations and continue to raise capital. Valuations do not collapse suddenly.

The AI investment bubble will burst imminently

The gigantic amounts of money flowing into AI companies in 2022-2026 is seen not to generate returns and AI company valuations collapse very soon.

AGI Safety Alignment Progress

Robust Solutions

Robust, verified alignment solutions are developed and thoroughly tested before AGI deployment, ensuring AI systems reliably pursue intended goals and values.

Minimal Safeguards

Minimal safeguards are implemented due to competitive pressures and technical challenges, leaving significant alignment problems unsolved at AGI arrival.

AGI Regulatory Environment

Global Coordination

Strong international governance frameworks emerge with coordinated policies, standards, and oversight mechanisms for AGI development and deployment.

Fragmented Landscape

Fragmented regulatory landscape develops with inconsistent national policies and AI development havens undermining global coordination efforts.

AGI Hardware Requirements

New Breakthrough Computing Technologies

AGI requires quantum, photonic, thermodynamic or other breakthrough computing technologies, creating additional technical barriers and development timelines.

Existing Architectures

AGI emerges through scaling existing semiconductor architectures and optimizing current computing paradigms without fundamental hardware breakthroughs.

AGI Economic Transition Speed

Gradual Integration

Gradual economic integration allows for smooth adaptation, retraining programs, and managed transition of workforce and business models.

Sudden Displacement

Sudden economic displacement occurs as AGI capabilities rapidly automate broad sectors, causing significant disruption and social adjustment challenges.

More ->