Answer-First Lead
Anthropic has warned that AGI could arrive by 2028 — and the company says the US must stay ahead of China in the AI race. The timeline represents a dramatic compression from forecasts just 12 months ago. Dario Amodei’s message isn’t just about technical capability: it’s about geopolitical positioning in a world where AGI arrival is measured in years, not decades.
🔍 THE BOTTOM LINE
AGI timelines have collapsed. Metaculus forecasters now put a 25% probability on AGI by 2029 and 50% by 2033 — down from a median of 50 years away just a few years ago. Anthropic’s 2028 warning is at the aggressive end of these estimates, but it’s not an outlier anymore. The question has shifted from “if” to “when” and “who gets there first.”
The China angle is critical. This isn’t just a capability announcement — it’s a geopolitical positioning statement. Anthropic is framing AGI development as a race where second place has meaningful downsides. That framing matters for policy, for investment, and for how we think about AI safety in a competitive landscape.
If you’re tracking AGI timelines: 2028 is now a serious estimate from a frontier lab. Not a prediction, but a planning horizon.
The 2028 Timeline
Anthropic’s warning, reported May 15, 2026, puts AGI arrival within two years. The context:
- Timeline: AGI possible by 2028 (2 years from announcement)
- Confidence: Framed as a warning, not a certainty — but serious enough to shape policy advocacy
- Messenger: Anthropic leadership, including CEO Dario Amodei
- Audience: US policymakers, with explicit China competition framing
This isn’t the first aggressive timeline from Anthropic. At Davos in January 2026, Amodei said: “We might be six to twelve months away from when the model is doing most, maybe all, of what software engineers do end to end.” That was about coding AGI specifically. The May 2026 statement is broader — general capability across domains.
Why Timelines Have Compressed
The shift from “50 years” to “2 years” isn’t random. Several factors:
1. Scaling Laws Hold
Every generation of models has followed predictable scaling curves. More compute + more data = better performance. The curves haven’t broken. If anything, they’ve accelerated with:
- Mixture of Experts — efficient scaling without proportional compute increases
- Better architectures — transformers refined, new attention mechanisms
- Synthetic data — models generating training data for other models
2. Capability Surprises
Models keep doing things researchers didn’t expect:
- Reasoning breakthroughs — GPT-5 disproving the Erdős conjecture (May 2026)
- Autonomous work — Qwen3.7-Max running 35 hours unsupervised on kernel optimization
- Security capability — Mythos finding 10,000+ vulnerabilities, clearing UK AISI’s cyber range
- Mathematical originality — Not just solving problems, but producing novel proofs
Each surprise resets expectations about what’s possible.
3. Investment Velocity
The money flowing into AI infrastructure is unprecedented:
- Anthropic: $30B+ raised at $900B+ valuation
- OpenAI: $25B ARR, $852B valuation, IPO filed
- Google Cloud: $200B commitment to Anthropic over 5 years
- Microsoft, Amazon, Nvidia: Hundreds of billions in AI infrastructure spending
This isn’t “let’s experiment” money. It’s “we believe AGI is imminent and we need the infrastructure ready” money.
The China Factor
Anthropic’s statement explicitly frames AGI as a US-China competition. This matters because:
1. Different Governance Models
- US: Private companies lead, government reacts (sometimes)
- China: State-directed, coordinated, long-term planning
- Implication: China can mobilise resources differently — potentially faster on some dimensions
2. Safety vs. Speed Tradeoff
- US labs: Publicly committed to safety testing, gradual deployment
- China: Less public constraint, potentially faster deployment
- Risk: If safety slows you down and your competitor doesn’t care about safety, you lose the race
This is the classic “race to the bottom” concern in AI safety. Anthropic’s statement is partly about building political support for safety measures that don’t cede ground to China.
3. Talent and Compute
Both matter for AGI:
- Talent: US still leads in AI research, but China produces more STEM graduates annually
- Compute: US has export controls on advanced chips to China, but China is investing heavily in domestic chip production
- Data: China has more internet users, different data access rules
The race isn’t just about who has the best researchers. It’s about who has the best researchers + the most compute + the most data + the most favourable governance environment.
What 2028 AGI Means
If Anthropic is right, we have two years. What changes?
For Policymakers
- Safety testing frameworks need to be operational, not theoretical
- International coordination becomes urgent — AGI doesn’t respect borders
- Economic preparation — labour market disruption could be rapid
- Military implications — AGI has obvious defence applications
For Companies
- Strategic planning horizon — 2 years is one product cycle for big tech
- Workforce planning — what roles are AGI-exposed?
- Infrastructure investment — compute, energy, data centres
- Competitive positioning — partner with AGI labs or build internally?
For Individuals
- Career planning — what skills are AGI-complementary vs AGI-substitutable?
- Financial planning — AGI arrival could reshape asset values, labour markets, entire industries
- Civic engagement — AGI governance will affect everyone, but most people aren’t paying attention yet
Counterarguments
Not everyone buys the 2028 timeline:
1. “AGI is a Moving Goalpost”
What counts as AGI? If it’s “human-level performance on all cognitive tasks,” we’re not there yet. If it’s “can do most knowledge work autonomously,” we might be closer. The definition matters.
2. “Scaling Hits a Wall”
Current progress assumes scaling laws continue. They might not. There could be fundamental limits we haven’t hit yet.
3. “Safety Slows Deployment”
Even if AGI-capable models exist by 2028, will they be deployed? Anthropic, OpenAI, and others have committed to safety testing. That could delay deployment even if capability arrives.
4. “China Isn’t as Far Along as the Rhetoric Suggests”
US export controls on advanced chips are biting. China’s AI progress is real but constrained. The “race” framing might be overstated.
For NZ
New Zealand isn’t in the AGI race — but we’ll be affected by its outcome:
- Economic exposure — NZ’s trading partners (US, China, Australia) will all be navigating AGI disruption
- Labour market — remote knowledge work is AGI-exposed. NZ has a lot of remote knowledge workers.
- Governance — NZ’s AI policy is still developing. AGI timelines compress the policy window.
- Security — AGI has implications for cyber defence, intelligence, regional stability
NZ doesn’t need to build AGI. But we do need to prepare for a world where AGI exists and our major partners are navigating its consequences.
📰 SOURCES
- Digit.in — “Anthropic warns AGI could arrive by 2028, says US must stay ahead of China in AI race” (15 May 2026)
- 80,000 Hours — “Shrinking AGI timelines: a review of expert forecasts” (March 2025)
- Novaknown — “Anthropic AGI Timeline: Why a Reddit Rumor Matters Less” (1 April 2026)
❓ FAQ
Q: Is 2028 a prediction or a possibility?
A: Anthropic framed it as a possibility — “could arrive by 2028” — not a certainty. But it’s a possibility from a frontier lab CEO, which makes it a serious planning horizon.
Q: What counts as AGI?
A: There’s no universal definition. Common criteria include: human-level performance across all cognitive tasks, ability to learn any intellectual task a human can learn, autonomous operation across domains. The definition matters for timeline estimates.
Q: Why does China matter?
A: If AGI is a race, and China is a competitor, then safety measures that slow US labs could cede ground. This creates pressure to deploy faster than might be ideal from a safety perspective.
Q: Should I believe this timeline?
A: Treat it as a data point, not a prophecy. Timelines have been wrong before. But the compression from “50 years” to “2 years” across multiple forecasters suggests something real is happening.
🔍 THE BOTTOM LINE (Reprise)
AGI timelines have collapsed from decades to years. Anthropic’s 2028 warning is at the aggressive end but not an outlier. The China framing matters: this is geopolitics, not just technology. Whether 2028 is right or wrong, the planning horizon has compressed. Policymakers, companies, and individuals should be acting accordingly.
