The Rundown AI / Articles / AI / Dario Amodei logs on to answer the critics

Dario Amodei logs on to answer the critics

PLUS: Hand off real work across your apps with Grok Bot

Zach Mink

August 17, 2026

Good morning, AI enthusiasts, and welcome to our 10,253 new readers. AI has a trust problem, and Anthropic CEO Dario Amodei doesn't think a marketing campaign can fix it.

Amodei just made a rare X appearance to answer claims that his AI safety talk is backfiring, framing the warnings as honesty rather than doom and pitching real breakthroughs in areas like medicine as the only route back for public opinion.

P.S. — We’re hiring for seven open roles at The Rundown! Click here to view the positions, apply, or refer the best person you know.

In today’s AI rundown:

Dario Amodei logs on to answer the critics
The Rundown Roundtable: Our AI use cases
Hand off real work across apps with Grok Bot
Doctor solves 22-year math problem with ChatGPT

LATEST DEVELOPMENTS

ANTHROPIC

✍️ Dario Amodei logs on to answer the critics

Image source: X / Images 2.0

The Rundown: Anthropic’s CEO Dario Amodei just chimed in on investor Gavin Baker’s X conversation with coworker Sholto Douglas, rejecting that his safety alarms have backfired and saying the only way AI wins trust back is by “actually curing cancer.”

The details:

Baker shared a rumor that Amodei envisions Anthropic one day becoming “the only private company in the world”, with Douglas calling it “completely false”.
Baker also argued Dario "has lost the argument" on AI rules and warned his risk talk is giving ammunition to anti-data-center campaigns in the U.S.
Amodei called locking down AI vs. spreading it wide via open models a “false choice”, saying Anthropic’s proposals slow big labs and spare small ones.
He also pushed back on owning AI’s negative image, saying "early glimmers” from biology and medicine work will shift sentiment more than marketing could.

Why it matters: Dario barely ever posts, so it’s noteworthy to see him jump in to defend both himself and the company, whose vibes in the AI world have taken some serious PR lumps lately. His reply lands on a trust issue that he sees no lab is marketing its way out of it, and in his view, only real medical breakthroughs can turn the tide.

TOGETHER WITH MEMOKET

⌚This wristband remembers everything

The Rundown: Memoket Gem is the lightest AI wristband on the market — a 0.4 oz wearable that records your conversations with one press and turns them into instant summaries, action items, and full context across every off-screen meeting.

With Memoket Gem, you get:

Hands-free capture as a wristband, Apple Watch clip, or pendant
One connected memory across conversations instead of 100 meeting notes
Seamless sync with ChatGPT, Claude, Google Calendar, and Notion
AI agents with full context from your real-life conversations via MCP

Pre-Order is open — get the device, your choice of accessory, and a free year of the app.

THE RUNDOWN ROUNDTABLE

💡 The Rundown Roundtable: Our AI use cases

Image source: Rishi @ The Rundown

The Rundown: The Rundown Roundtable is a weekly feature where we poll members of The Rundown staff about how we use AI in our work and daily lives.

Rishi, Growth: I use Claude Cowork and Meta MCP to map my existing ad creative data and find winning combinations.

I connect Cowork to Meta MCP and Google Drive, and place the brief for every ad I published in the last 365 days into one Drive folder that includes the script, format, offer, and persona it was written for. I then prompt Claude to work through the folder and build a spreadsheet that tags each ad by those attributes, have it pull each ad’s spend and CPA through Meta, and turn the spreadsheet into a flowchart.

Once the data is mapped, I can identify one concept that wins more often than the others, has absorbed more spend, and still maintains an efficient CPA. That is the winning concept, and it is where I should focus 60% to 80% of my production time.

Shubham, Editor: My car recently lost a small part near the bonnet. I didn’t know what it was called, and even the official workshop couldn’t help because the model was no longer in production.

I gave ChatGPT my car details and a photo showing where the part was missing. It identified the part, helped me cross-check the specifications and compatible part numbers, and found sellers that had it available. I found the right part and had it delivered within a week.

AI TRAINING

🙇 Hand off real work across apps with Grok Bot

The Rundown: In this guide, you will learn how to turn Grok Bot into a focused team that works across your apps, including how to set it up, which agents to create, and how to create routines to complete work while you sleep.

Step-by-step:

Download Grok Bot, sign in with Cursor, and start the seven-day trial. Create your first Chief of Staff bot, and keep the workspace focused on one project
Next, connect the apps (Gmail, Drive, Calendar, Notion, etc.) you want your Grok Bot to use
Ask: “Review my connections, build a profile of my workstreams/projects, and suggest the three most useful agents, automations, and plugins I should add”
Now tell the Chief of Staff to build its first deliverable or report for you

Pro tip: Tell the Chief of Staff to set up a daily debrief routine for each Grok Bot so you can see what they worked on. Then use the full video guide to build the setup.

PRESENTED BY SANDBOXAQ

⚡ Prove your agents in two commands

The Rundown: Flint AI's open-source CLI helps engineering teams prove their AI agents are reliable before they ship. Find every agent in your codebase, then red-team each one for jailbreaks and data leakage. Make AI earn it.

Run locally:

flintai scan finds every agent in your codebase
flintai eval red-teams before ship
Score reliability before deploying

pip install flintai-cli or get started at flintai.dev.

AI & MATH

🧮 Doctor solves 22-year math problem with ChatGPT

Image source: Alextownsend.net

The Rundown: Beijing neurosurgery resident Shanmu Jin proved Crouzeix’s Conjecture, a matrix problem unsolved since 2004, with GPT-5.6 Sol surfacing the result after a 16-hour autonomous session in ChatGPT Work.

The details:

Jin reached the problem through brain-ultrasound research, teaching himself advanced math after undergraduate geology and medical training.
His setup denied the model internet access and used swarms of subagents against each other, a recipe borrowed from one of OpenAI's math prompts.
Formal peer review is pending, but Cornell’s Alex Townsend, Anne Greenbaum, and the problem’s own author have all verified the proof.

Why it matters: Math is having a wild AI summer, and it's not only the field's own experts landing the breakthroughs. That might sting for professional mathematicians, but their role is shifting rather than shrinking. When a proof can come from anyone with a ChatGPT subscription, verifying it ends up becoming a scarce, in-demand skill.

QUICK HITS

COMMUNITY AI WORKFLOW OF THE DAY

▸ Patrick built an AI speech coach for his public speaking class

Today’s workflow comes from reader Patrick Loebs:

“I teach and direct a general education public speaking class. I've built the course around critical inquiry, critical thinking, and public advocacy of a localized problem.

With ChatGPT, I built an AI tool. It begins with a general question: 'what is your topic and what is the problem?' Students respond with a one-word answer, such as 'poverty,' 'crime,' or 'the high cost of education.' The AI pushes back on these. It consistently asks, 'what do you mean by x?'

The AI does not create speeches, outlines, or find sources for students. Instead, it asks questions that help them refine the direction of their speeches. Whenever possible, the AI also identifies alternative viewpoints from sources traditionally associated with the student's own perspective. This helps students recognize that people on the same political, religious, or ideological side do not always agree on every topic.”

See Patrick’s workflow here. How do you use AI? Tell us for a chance to be featured.

🛠️ Trending AI Tools

🚀 GLM-5.3 - Z AI’s open model with strong coding, agentic, cyber capabilities
🔊 Pika Audio - Pika’s audio models spanning music, speech, SFX, soundtracks
🖥️ Computer History - Turn Mac activity into memory for ChatGPT and Codex
⚡️ Gemini Flash 3.7 - Google’s new fast, cheap and capable mid-class model

📰 Everything else in AI today

China’s Z AI rolled out GLM-5.3, claiming the strongest open-source coding model and top scores on certain cyber benchmarks, with weights coming in two weeks.

OpenAI introduced Computer History, a new opt-in feature that logs clicks and typing to give ChatGPT and Codex a memory of recent work.

The Information reported on Dario Amodei’s wife Cami Clark, including her past ventures and an attempt to secure Jeffrey Epstein’s backing for an adult film company.

Pika Labs released four AI audio models covering soundtracks, music, sound effects, and speech, with the startup claiming the family runs up to 20x cheaper than its rivals.

Anthropic shared more details on Claude watermarking plans, saying the marks add no cost or hidden characters, and won’t carry any info traceable to users/organizations.