AI

What changed

Recent signals

• Frontier-AI risk became an explicit object of coordinated international governance through the Bletchley Declaration.retrospective mapping
  supported inference
• Frontier-model governance moved from mainly voluntary/private commitments toward formal executive requirements and provisional model-specific legislation.retrospective mapping
  supported inference
• Long-context, multimodal and tool-capable model interfaces expanded across multiple providers and developer platforms while some frontier variants remained preview or unavailable.retrospective mapping
  supported inference
• Agentic cyber tooling has moved beyond exposed infrastructure into documented adversarial operational use, while the degree of autonomous contribution and actor attribution remain unresolved.
  supported inference
• Frontier-model governance increasingly used concrete voluntary red-team, security, transparency and capability-threshold safeguards while remaining largely private or nonbinding.retrospective mapping
  supported inference
• Conversational generative AI began appearing as an operating-system-level interaction surface through the Windows Copilot preview rollout.retrospective mapping
  supported inference
• Public assistant interaction began expanding from text into image and voice in late Q3, with rollout incomplete at the quarter cutoff.retrospective mapping
  supported inference
• AI governance became an explicit subject of multilateral coordination among G7 governments, with emphasis on trustworthy AI, standards and interoperability.retrospective mapping
  supported inference
• Generative AI interaction spread further across established software surfaces and geographies, including broader Bard access and increasingly integrated provider products.retrospective mapping
  supported inference
• AI governance expanded from frameworks and enforcement toward accountability mechanisms, compliance-driven product adaptation and active legislative negotiation.retrospective mapping
  supported inference
• NYC frontier-AI oversight progressed from planned hearing to completed public testimony and a scheduled next legislative step, while enacted technical-control requirements remain absent.
  supported inference
• AI governance became more operational through finalized risk-management guidance and direct regulatory enforcement against a deployed generative-AI service.retrospective mapping
  supported inference
• Existing data-protection governance directly constrained a deployed generative-AI software service through a regulator-imposed temporary processing restriction in Italy.retrospective mapping
  observed fact
• Conversational, speech and generative model capabilities increasingly became directly programmable developer infrastructure through APIs and prototyping platforms.retrospective mapping
  supported inference
• GPT-4 established a materially stronger frontier-model capability state in Q1, while public access remained gated and image input was not yet generally available.retrospective mapping
  supported inference
• Public conversational generative AI expanded from Q4's breakout ChatGPT interface into multiple provider, search and browser surfaces, while access remained preview-limited in several products.retrospective mapping
  supported inference
• U.S. AI-policy coordination has become more tightly institutionally coupled to the national-security/intelligence apparatus through the DNI's formal AI-czar and task-force role.
  supported inference
• AI governance became institutionally more concrete as the EU Council adopted a negotiating position on a horizontal risk-based AI regulatory framework, while deployment and safety controls remained heterogeneous.retrospective mapping
  supported inference
• A general conversational AI interface became directly and freely accessible as a public research preview, making multi-turn instruction-following interaction a first-class user surface rather than requiring an API, specialist workflow or narrow research demo.retrospective mapping
  supported inference
• Instruction-following and dialogue behavior became more deliberately shaped through scaled instruction finetuning and RLHF-based conversational fine-tuning, with public interaction exposing capability and reliability limits.retrospective mapping
  supported inference
• AI-generated outputs became materially embedded as reusable shared scientific infrastructure rather than only individual model predictions.retrospective mapping
  supported inference
• AI governance became more operational through structured risk-management guidance, explicit dialogue-agent rules, human feedback and deployment safeguards, while remaining largely nonbinding.retrospective mapping
  supported inference
• Generative-image capability moved from controlled previews toward materially broader public use through both hosted mass access and downloadable model weights.retrospective mapping
  supported inference
• Model access broadened materially from controlled frontier research access toward downloadable or openly reusable model artifacts across language, translation, image generation and speech, while licenses and hardware requirements remained real constraints.retrospective mapping
  supported inference
• Model-generated code moved from Q1 research demonstration toward broadly available, editor-integrated developer assistance.retrospective mapping
  supported inference
• Public and private governance moved closer to the technical stack through AI-risk measurement work, semiconductor/export-control coordination and controlled model deployment/release practices.retrospective mapping
  supported inference
• Model interfaces broadened materially beyond text toward interleaved vision-language understanding and large-scale text-guided image generation in real user workflows.retrospective mapping
  supported inference
• By the end of Q1 2022, the clearest stack coupling ran through research, models, software, compute, chips, networks, manufacturing and governance; the selected evidence does not yet support persistent agents or agent permissioning as independent operational layers.retrospective mapping
  supported inference
• AI risk management was becoming formalized as a structured institutional governance process spanning the AI lifecycle.retrospective mapping
  supported inference
• Large language models were becoming capable of producing nontrivial program solutions, establishing an early direct dependency path from model capability into software work.retrospective mapping
  supported inference
• Prompt structure was beginning to function as a software-level capability surface that could elicit additional reasoning behavior from fixed model weights.retrospective mapping
  supported inference
• Dedicated AI supercomputing was becoming a distinct operational layer, with model progress increasingly dependent on large accelerator fleets, high-bandwidth networking and storage.retrospective mapping
  supported inference
• Human-feedback post-training moved from research technique into default production model behavior in a major commercial API.retrospective mapping
  supported inference
• Frontier model releases are increasingly using phased access for high-end cybersecurity capability, coupling model competition with deployment-governance processes.
  supported inference
• Agent safety is moving beyond voluntary engineering controls into formal enforcement, evidence demands and potential liability scrutiny in the United States.
  supported inference
• AI competition policy is increasingly turning on access to incumbent data infrastructure, while privacy and security are becoming central countervailing claims.
  supported inference
• AI policy is becoming a more explicit electoral and party-agenda issue in the United States.
  supported inference
• The AI stack is becoming more tightly coupled across models, agents, permissions, compute, chips, software, networks, energy and governance.
  supported inference
• Physical and spatial AI are moving into major chip-vendor strategy, tightening the link between AI models, robotics, simulation and hardware roadmaps.
  supported inference
• Frontier-level AI capability is being pushed into faster and lower-cost model tiers, reducing the cost barrier for sustained professional and agentic workloads.
  supported inference
• Persistent personal agents are becoming a first-class product category rather than isolated one-shot task tools.
  supported inference