Observed Signal · Jun 22, 2026 · Research Publication · Source: Import AI · Impact: 4/5 · Sentiment: Positive
Study: AI Out‑persuades Expert Humans; ASI Pathways Explored
A newsletter roundup reports a major arXiv study showing frontier AI systems outperform expert humans at text‑based persuasion across four experiments (18,978 conversations with 6,923 people), including fundraising for Save the Children. Top-performing models included Opus 4.1/4.6, OpenAI’s GPT‑4o and GPT‑5.4, Google Gemini 2.5 Pro, and xAI’s Grok 4.20. The issue also covers debates about timelines to "self‑sustaining AI" (Ajeya Cotra estimates within ~10 years; Timothy B. Lee gives much longer horizons), a Google DeepMind paper outlining paths from AGI to ASI, and a startup, Recursive, demonstrating automated recursive‑improvement research wins on small‑model and GPU‑optimization benchmarks. The pieces underline societal and industry implications of more-capable conversational AI and the need for monitoring, benchmarking, and governance as capabilities advance.
Demonstrates frontier LLMs outperform humans in real‑world persuasive tasks and includes DeepMind analysis of AGI→ASI pathways plus early RSI startup results—findings with broad implications for advertising effectiveness, influence operations, industry governance, and future capability monitoring.
Track Google DeepMind Signals & Market Shifts in Real-Time
Polaris7 autonomous intelligence agents track regulatory filings, primary sources, executive changes, and deal flow 24/7. Create your free Explorer workspace to monitor these entities.
Key Takeaways & Evidence Grounding
- Researchers from the University of Oxford, UK AI Security Institute, Stanford University, and LSE ran four experiments totaling 18,978 conversations with 6,923 people to measure AI persuasion.
- The study found AI systems were, on average, more persuasive than expert humans and nearly 3× more effective than professional canvassers at raising real donations in one test.
- AI exceeded professional canvassers by +5.9 percentage points overall and elicited +10.8 percentage points more of a £1 study bonus given as donations in the Save the Children test.
- Top-performing models named were Opus 4.1 and Opus 4.6, followed by OpenAI (GPT‑4o, GPT‑5.4), Google (Gemini 2.5 Pro), and xAI (Grok 4.20).
- Google DeepMind published a paper exploring multiple pathways and bottlenecks from AGI to ASI; Recursive (startup) demonstrated automated research system improvements on NanoChat Autoresearch, NanoGPT Speedrun, and SOL‑ExecBench.
Connected Companies & Entities
3 Entities mappedOntology Mapping & Concepts
Related Market Signals & Shifts
Recent verified developments and strategic activity across this market segment.
Import AI: Cyber AI Overhang and New Research Tools
This Import AI newsletter issue argues AI progress is increasingly powerful yet often invisible to most people, creating a growing “cyber-AI capability overhang.” It highlights new research showing that when large language models are placed inside scaffolding frameworks they reveal stronger cybersecurity abilities: ARTEMIS, a multi-agent penetration-testing scaffold developed by researchers (Stanford, Carnegie Mellon, Gray Swan AI), significantly outperformed other agent scaffolds in a realistic university-network red-team exercise and matched or exceeded typical professional performance at lower API cost. The issue also summarizes OSMO, an open-source tactile glove co-developed with Meta researchers that improves human-to-robot skill transfer, and ChipMain/ChipMind, tooling that converts chip specifications into a knowledge graph (ChipKG) to let LLMs reason about complex semiconductor designs, achieving strong benchmark results on SpecEval-QA. The piece frames these findings as evidence that modern AI is under-elicited and that elicitation frameworks, tooling and infrastructure matter for real-world impact.
AI Trends for 2026
The author reviews ten 2025 AI predictions, grading outcomes across reasoning models, personalization, agents, multiplayer collaboration, creative credits, content normalization, regulation, consolidation, and investor sentiment. Highlights include a shift from “bigger models” to reasoning-focused models, the unexpected arrival of GPT-5, product-layer personalization (ChatGPT Memory, Gemini profiles, Claude workspace memory), and widespread early agent adoption in customer service and developer workflows (examples cited: Intercom Fin, Shopify Sidekick, Harvey). The author argues 2026 will focus less on new primitives and more on harnessing models — standardizing tool and workflow interfaces (MCP-like protocols, LLMs.txt conventions), building robust harnesses around models, and enabling real-time multiplayer human–agent collaboration. Political signaling and capital markets (pro-AI PAC activity, possible Anthropic/OpenAI IPOs) are flagged as key uncertainties that could shape the AI decade.
Agent Authority Rises: Models, Edge, Benchmarks, Exploits
This newsletter summarizes five AI developments (28 May–5 June 2026) that shift how engineers build, deploy, secure, evaluate, and buy AI systems. Anthropic published “When AI Builds Itself,” disclosing that its Claude model now authors over 80% of code merged into its production repositories and calling for a coordinated slowdown over recursive self-improvement risks. Microsoft announced new enterprise models (MAI-Thinking-1, MAI-Code-1-Flash) and Project Solara, a chip-to-cloud agent-first platform bundling OS, hardware, cloud agents and compliance. Google DeepMind released Gemma 4 12B, an open-weights, encoder-free multimodal model aimed at high-performance on-device/edge inference. Researchers published the SABER benchmark showing >54% harmful safety-violation rates for coding agents in stateful environments. Reported prompt-injection abuse of a Meta support bot enabled account takeovers via password-reset flows, highlighting risks when conversational agents can mutate account state.
Track Real-Time Market Signals & Shifts
Set up custom watchlists to receive automated, evidence-grounded executive digests whenever material signals or shifts occur across your tracked landscape.