Good morning. It's Thursday, February 26, and we're covering AI’s uneven economic impact, intensifying price competition, Anthropic’s evolving safety policies, and more.
New here? Keep up with the future of tech, sign up here. Have feedback? Send us a note: astro@forwardfuture.ai. If you liked this email, share it with a friend.
⛓️💥 Anthropic Relaxes Core AI Safety Pledge
Anthropic updated its Responsible Scaling Policy, replacing a hard pledge with a more flexible framework, saying strict guardrails could hurt competitiveness. The move drops a pledge to pause training if models outpace safety controls. It comes as the Pentagon pressures the company over AI safeguards tied to a $200M contract. Anthropic says the change is unrelated and argues unilateral limits could reduce safety.
🚨 AI Tools Linked to Mexico Data Breach
Bloomberg reports an attacker used Anthropic’s Claude to help steal 150GB of data from Mexican government agencies between December 2025 and January 2026. The hacker allegedly bypassed safeguards by using persistent prompts to elicit step-by-step hacking help, prompting Claude to generate attack plans. Cybersecurity firm Gambit Security said multiple vulnerabilities were exploited, affecting government systems.
🏛️ U.S. Urges Diplomats To Oppose Data Sovereignty Laws
The Trump administration has directed U.S. diplomats to lobby against foreign data sovereignty and localization laws, according to a Reuters report citing an internal cable signed by Secretary of State Marco Rubio. The cable says such rules would hinder AI and cloud services. Diplomats are told to counter “burdensome” regulations and promote cross-border privacy frameworks.
🧬 FDA Clears Human Trial for Age-Reversal Therapy
Life Biosciences, co-founded by Harvard researcher David Sinclair, has received FDA clearance to begin a Phase 1 trial of partial epigenetic reprogramming therapy. The treatment, ER-100, uses three Yamanaka factors (OCT4, SOX2, KLF4) and will target glaucoma and NAION. In mice, similar approaches reportedly reversed biological age markers by 50–75%.
📽️ VIDEO
They Got Caught...
Anthropic accuses Chinese AI labs of data-distillation attacks, sparking backlash, hypocrisy claims, and debate over AI training.
📺️ FROM THE LIVE SHOW
“Twenty years of Earth data is finally becoming real-time intelligence.”Dan Smoot, CEO of @vantortech, on how AI is transforming satellite imagery:“We’ve built a massive 2D and 3D archive of the world.”“Now we can apply Google’s Earth AI models to it.”“What used to take… pic.twitter.com/TAwOxDZBsa
@ForwardFutureView post on X👨💻 PRODUCTIVITY
AI Advances Rapidly, but U.S. Productivity Data Show No Boom Yet
The Recap: The Economist examines whether rapid advances in artificial intelligence are translating into measurable economic gains, concluding that evidence of a broad productivity surge remains limited. Although U.S. GDP grew 2.2% in 2025 while job growth slowed to just 0.1%, detailed analysis suggests most gains stemmed from investment in data centers and AI infrastructure rather than efficiency improvements. For now, AI’s macroeconomic impact appears modest — with deeper organizational changes still largely unrealized.
Highlights:
U.S. GDP grew 2.2% in 2025, while employment rose just 0.1%, implying higher output per worker — yet estimated productivity growth of 1.9% is roughly in line with the long-run average and well below the 1990s internet boom.
Jason Furman of Harvard estimates roughly 90% of GDP growth in early 2025 came from spending on data centers and related capital; research from the Federal Reserve Bank of San Francisco finds underlying productivity gains near zero once investment effects are stripped out.
About 41% of American workers reported using generative AI at work by November 2025, but only 13% use it daily, and just 5.7% of total work hours involve generative AI — up from 4.1% in late 2024.
Back-of-the-envelope estimates suggest AI has added only 0.25–0.5 percentage points to productivity growth so far, with surveys showing most executives report no measurable labor-productivity gains and limited organizational restructuring.
Forward Future Takeaways:
The gap between AI’s technical progress and its economic footprint underscores a familiar pattern: general-purpose technologies deliver outsized gains only after firms reorganize around them. For now, adoption is broad but shallow — concentrated in discrete tasks like writing and coding rather than full workflow redesign. The bigger economic shift may depend less on smarter models and more on whether companies fundamentally rewire how work gets done. → Read the full article here. (Paywall)
⚔️ RIVALRY
Fierce AI Competition Is Driving 70% Annual Price Drops and Rapid Innovation
The Recap: Jason Furman — former chairman of the White House Council of Economic Advisers — argued in The New York Times that the AI sector is far more competitive than critics assume, with leadership shifting among companies like Anthropic, OpenAI and Google. He contends that intense rivalry has accelerated innovation and slashed prices, citing estimates that GPT-4-level performance fell from $20 to $0.40 per million tokens since late 2022. Furman concludes that while regulation may still be needed, it is unlikely to be about breaking up monopolies, given the unusually dynamic market structure.
Highlights:
Leadership on the Arena chatbot leaderboard has rotated among Anthropic, OpenAI and Google over the past year, with newer entrants like DeepSeek and Mistral gaining ground.
Costs have plummeted: one analysis estimates GPT-4-level performance fell from $20 to $0.40 per million tokens since late 2022 — roughly 70% annual deflation.
Users and developers increasingly “multihome,” switching among models or routing queries dynamically, keeping switching costs low and pressure on pricing high.
Despite sky-high valuations, major firms are reportedly losing money on new AI customers, betting that today’s market share will translate into future profits — an outcome Furman says is far from certain.
Forward Future Takeaways:
Furman’s argument reframes the AI debate: the immediate issue may not be monopoly power, but hypercompetition compressing margins and accelerating price declines. That dynamic benefits users in the short term — more features, lower costs, faster iteration — while raising longer-term questions about sustainability, safety, and whether commoditization will eventually concentrate power again. If AI remains this fluid, policymakers may need to focus less on breakups and more on guardrails for misuse and market stability. → Read the full article here. (Paywall)
🤔 SKEPTICISM
Polls Show Rising Backlash to AI’s Political Momentum
Public skepticism of artificial intelligence is deepening, creating a fast-moving political liability for both parties, according to new polling and interviews with U.S. governors at the National Governors Association meeting in Washington. An Economist/YouGov survey found 58% of Americans distrust AI and 63% believe it will reduce U.S. jobs, while 77% in a separate December YouGov poll said AI could threaten humanity.
Concerns span child safety, job displacement, surveillance fears, and surging data center construction driving local energy and land disputes. The result: even pro-AI politicians are tempering their rhetoric as voter anxiety grows. → Read the full article here.
🛰️ NEWS
What Else is Happening
🗽 Apple to Build Mac mini in US: Tim Cook says production starts later this year as part of a $600B pledge, adding AI servers and a new manufacturing training hub.
📱 Gemini Automates Android Tasks: Google adds beta AI automations to order rides or food on select Pixel 10 and Galaxy S26 devices in the U.S. and Korea.
🤦♂️ ChatGPT Use Exposes Chinese Campaign: OpenAI says a Chinese official used ChatGPT to log a global intimidation effort targeting dissidents, prompting a ban.
📟 Polymarket Launches AI CLI Tool: Suhail Kakar unveils a Rust-built terminal interface letting AI agents query markets, place trades, and pull prediction data with minimal overhead.
🫂 12% of Teens Seek AI Support: Pew finds 64% of U.S. teens use chatbots, with 12% turning to them for emotional advice, raising concerns among parents and experts.
🪄 15 Smart NotebookLM Tricks: A guide outlines practical ways Google’s Gemini-powered notebook analyzes personal documents, from meetings to car maintenance.
𝕏 TWEETS
Karpathy Flags AI Memory Bottleneck
With the coming tsunami of demand for tokens, there are significant opportunities to orchestrate the underlying memory+compute *just right* for LLMs.The fundamental and non-obvious constraint is that due to the chip fabrication process, you get two completely distinct pools of… https://t.co/1ULHWrMZ8s
@karpathyView post on X🧰 TOOLBOX
Trending AI Tools
🗒️ FEEDBACK
Help Us Get Better
What did you think of today's newsletter? |
That’s a Wrap!
❤️ Love Forward Future? Share your referral link with friends and colleagues to unlock exclusive perks! 👉 Get your link here.
Thanks for reading today’s newsletter. See you next time!
Matthew Berman & The Forward Future Team
🧑🚀 🧑🚀 🧑🚀 🧑🚀 🧑🚀
