In partnership with
October 19, 2026 · READ ONLINE
Keeping you ahead in AI.
THE DAILY BRIEFING / 4 MIN READ
Welcome back, AI enthusiasts.
Anthropic is pledging 5% of itself to public causes, and people who lose their jobs to AI could be among those it helps. That stake could be worth $100 billion at a $2 trillion IPO valuation, though there’s no worker payment program yet. Coming ahead of its planned listing, the promise could earn Anthropic some goodwill while leaving people waiting to learn what help they might actually receive.
IN TODAY'S ISSUE
01 Anthropic Pledges 5% to Public Benefit Fund Ahead of IPO
02 Google Helps AI Agents Get Better With the Same Model
03 Anthropic Wants Courts to Decide When AI Stops
A FRESH LOOK
You might have noticed something different about today's newsletter! We've officially rolled out our new design, giving Evolving AI Insights a fresh, cleaner look.
While the design has changed, our mission remains the same: bringing you the most important AI news, insights, and tools to keep you ahead of the curve. We hope you like the new look. Now, let's dive into today's AI news!
01 / PUBLIC AI FUND
Evolving AI: Anthropic is pledging a stake potentially worth $100 billion to public causes including possible support for people who lose their jobs to AI.
KEY POINTS
The 5% share pledge needs shareholder approval, with the valuation estimate based on a possible $2 trillion IPO.
The board can add up to 2% more shares for every 50% rise in company value above an undisclosed baseline.
Together with separate founder and employee pledges, the fund’s allocation would bring total public benefit commitments to roughly 20% of Anthropic’s shares.
DETAILS
Anthropic is creating the Public Benefit Fund to share some of the wealth its AI generates with the wider public. The fund could make large donations to nonprofits or give directly to specific groups, including workers affected by AI. Anthropic will own it as a limited liability company to allow different forms of giving, with fuller governance details and plans for who receives support still to be announced.
WHY IT MATTERS
Anthropic’s June Economic Policy Framework proposed income support and ways to share AI wealth, while July’s agenda for its $200 million Economic Futures Research Fund outlined research into how those approaches could work. The new share pledge could help finance those ideas while building goodwill ahead of an IPO, as some investors lower their valuation expectations from $2 trillion to around $1.5 trillion. There is no universal income commitment or announced worker payment program, and Anthropic will own the new fund. Its lasting value will depend on whether people affected by AI receive support they can rely on and have a say in how it is provided, long after the attention around the IPO has passed.
From invisible to first in ChatGPT.
Bob's Your Uncle has spent twenty years helping challenger brands win their categories. But when buyers asked ChatGPT about their own category, the agency didn't show up.
With ChatFeatured, their AI visibility went from 7% to 22% in 90 days, and they went from invisible to first in ChatGPT.
They're not the only ones:
Vessel: AI visibility up 2,200%+
SCOJO: AI visibility up 160%
WikiSleep: brand mentions up 113%
ChatFeatured finds the questions where AI leaves you out, writes the content that gets you cited, and reports the lift every week. Your team decides what ships.
Most customers see their first citations inside 7 days.
02 / AI HARNESS
Source: The Decoder
Evolving AI: Google has released Harness² to let AI agents test and refine their own working setup without developers having to design every change.
KEY POINTS
Agents can adjust 8 parts of their setup, including instructions, memory and tool guidance, while the underlying model stays unchanged.
One refinement step raised GLM-5’s WorkBuddy-Bench score from 64.2 to 75.7, beating Gemini 3.5 Flash’s original setup in the same test.
The system saves revised setups and lessons in a library that helps guide how it approaches future tasks.
DETAILS
Google’s Harness² is an open-source framework that lets AI agents adapt how they work to the task they are given. It proposes changes to their setup, tests different versions on the same task and combines what it judges useful. It works without an answer key, comparing the agents’ actions and outputs to assess the changes. Those judgments can still be wrong, and the extra test runs add costs. The research code is available under Apache 2.0 for developers to build on.
WHY IT MATTERS
Google’s research suggests we could get more useful AI from models we already have. Its earlier ReasoningBank project helped agents remember lessons from past attempts, and Harness² lets them use that experience to change how they work. In one example, an agent improved after it began checking source files against each other and reviewing its finished work. If those habits carry over to new tasks, developers could spend less time fixing the same problems. People would still need to check the agent’s changes, but the hope is that this becomes less demanding over time. That would make it easier to give an agent more responsibility without spending more of the working day supervising it.
TOGETHER WITH DATADOG
Evolving AI: Your Guide to Building Secure AI Applications.
As AI adoption accelerates, new attack surfaces are emerging across infrastructure, supply chains, and model interfaces. Datadog’s AI Security Best Practices Guide breaks down how to secure:
The underlying components that host and run AI applications
The software and data that an AI application uses to operate
The entry points and business logic that enable a user to interact with an AI application
This guide provides actionable strategies to help teams strengthen AI security without slowing innovation.
03 / AI POLICY
Source: Bloomberg
Evolving AI: Anthropic has proposed an AI safety bill that would require independent reviews of powerful models and give courts the final say on when the government can shut them down.
KEY POINTS
Shutdowns would require an imminent risk of more than $1 billion in damage or the death or serious injury of more than 50 people.
The Commerce Department would need a federal judge’s approval to halt a model, with other federal agencies barred from ordering suspensions.
Independent evaluators would have 72 hours after completing an assessment to report an imminent risk covered by the bill.
DETAILS
Anthropic’s draft would require AI developers to test their most powerful models and submit to outside evaluations. It targets biological weapons, cyberattacks and AI escaping its developer’s control. Labs could pay evaluators as long as payment did not depend on their findings. The proposal would also override certain state laws, making the strength of its federal protections especially consequential. It remains a draft being pitched to lawmakers.
WHY IT MATTERS
Anthropic’s proposal could protect AI companies and their customers from arbitrary government shutdowns. The harder judgment is how much evidence officials should need before they can act on a safety warning. A test could reveal dangerous behavior without establishing when it might cause harm or how many people could be affected. Congress will need to account for that uncertainty, especially if federal rules replace some state protections. A fair process should let companies challenge the evidence while allowing temporary restrictions when credible warnings need urgent investigation, so people remain protected while the risks are being assessed.
04 / HUMAN OR MACHINE?
One is real. Trust your eyes?
Two images. One photograph. Click the one you think is real.
Image 1
Image 2
05 / QUICK HITS
👶 Anthropic’s own report found junior staff fell from 27% to 9% of its technical hires as AI agents now log six times the hours of the humans directing them.
🔑 OpenCode confirmed a breach it blames on a “distillation attack” that exposed saved OpenAI, Anthropic and Google API keys for 1.4% of Console users, with fraudulent charges reported.
🧮 OpenAI, Anthropic and the Simons Foundation committed $15M to mathematics to help researchers verify AI-assisted discoveries and establish shared standards.
🏦 Citi and Bank of America joined JPMorgan in blocking third-party AI agents from logging into customer accounts, fearing agents will sweep deposits to higher-yield banks..
06 / TRENDING AI TOOLS
🚀 Fyxer - email assistant that organizes your inbox, writes draft replies in your tone, and takes meeting notes*
✅ Afterparty - Turns meeting transcripts into emails, invites and follow-up tasks.
🎬 Rushilo - Edit videos by telling Claude Code or Codex what to change.
🎤 Toby - Runs realistic mock interviews and tells you what to improve.
*PARTNER LINKS
BEFORE YOU GO
Worth your four minutes?
Help us make the next edition better.
See you tomorrow,
The Evolving AI team
