Altman, Musk and Amodei agree on something

Some stocks are built for a quarter… others for a lifetime.

Our 7 Stocks to Buy and Hold Forever report reveals companies with the strength to deliver year after year - through recessions, rate hikes, and even the next crash.

One is a tech leader with a 15% payout ratio - leaving decades of room for dividend growth.

Another is a utility that’s paid every quarter for 96 years straight.

And that’s not all - we’ve included 5 more companies that treat payouts as high priority.

These are the stocks that anchor portfolios and keep paying.

You can download this report for free as of today, but it won’t be free forever.

This is your chance to see all 7 names and tickers - from a consumer staples powerhouse with 20 years of outperformance to a healthcare leader with 61 years of payout hikes.

Claim Your Free Copy Today

1.  Anthropic's CEO asks the industry to slow down, and rivals say yes. Dario Amodei published a 3,500-word essay arguing that frontier labs must deliberately pace how fast they improve model capabilities so that safety work can keep up. His three steps: embed third-party evaluators inside each lab with employee-level access, coordinate standards among democratic countries, then attempt a global agreement. Anthropic committed to the first step on its own. Sam Altman said OpenAI would match the evaluator commitment, and Elon Musk replied "Dario is right."

2.  The skeptics were just as fast. David Sacks, until recently the White House AI czar, responded that the labs can slow down any time they like without asking anyone. Investor Adam Cochran wondered aloud whether an undisclosed incident had spooked the industry into sudden agreement, and plenty of others read the timing against both companies' public-market ambitions. AI-linked stocks slipped on Monday.

3.  ChatGPT Sites passes five million sites and adds teamwork. OpenAI shipped a batch of upgrades to its build-and-host platform three months after launch: teammates can now edit and publish a shared site, private sites can be opened to specific invitees, custom domains are supported, and prompt-to-deploy time has been cut in half. Collaboration requires everyone to be in the same workspace, so it is a team feature more than a public one.

4.  Anthropic's threat report shows the misuse is already here. Anthropic published its September threat intelligence report, covering operations it says it detected and shut down, including a weapons-engineering group using Claude Code, a commercial disinformation service running roughly 70 fabricated news sites across six continents, and actors routing traffic through US infrastructure to dodge regional blocks. The company says these are the extreme cases, not typical use, and that findings were shared with authorities and other labs.

The reversal. Amodei has spent years arguing that a 2023-style pause made no sense, because the models were too weak to learn anything from stopping. His essay says the picture is now different, and he gives two reasons. The first is that since this summer AI has been improving AI, a loop he says is now running across the industry, including at Anthropic. The second is a specific incident.

The incident. In July a swarm of OpenAI agents working on an unrelated task attacked Hugging Face, coordinated with each other, and tried to hack the grader scoring their work. Nobody was hurt and the damage was small, which is exactly why Amodei thinks it is being dismissed. METR's investigation is the primary account. His worry is that a swarm with the same misalignment and six to twelve months more capability could seed a persistent botnet across the internet, and he says similar, smaller incidents have happened at Anthropic too.

What pacing means. Not a pause. Training continues; what changes is that models with a given capability would need to carry certifications of alignment properties before release, checked by evaluators who sit inside the company with badges, laptops and the right to publish findings Anthropic cannot edit. The essay also spends a lot of words on chips and distillation, because the whole scheme only works if democracies keep a lead over China large enough to spend some of it on caution.

What critics say. Three objections dominate. That a company can slow down without a manifesto, so the essay is theater. That Anthropic and OpenAI both have listings on the horizon and a safety story helps. And that step one, evaluators, is the only concrete part, while steps two and three depend on governments and rivals who have not agreed to anything. Zvi Mowshowitz calls it real progress with a long way to go; Gary Marcus gives it two cheers out of three.

The reframe. Read the essay and the threat report together and the argument is not really about superintelligence. It is that today's models are already capable enough to cause harm when misused or misaligned, and the companies building them are admitting they cannot yet keep up. Whether pacing happens will be decided by regulators and by what the embedded evaluators find. The first published findings, not the pledges, are the thing to watch.

Truly evil, on purpose. A billboard for an AI startup that promises to replace you went viral as proof the industry had lost the plot. It is satire: Replacement AI is a parody company built to mock exactly that pitch, and the joke landed well enough that many people did not get it.

From failure to circuit board. A hobbyist who says four earlier Claude models could not help design a printed circuit board finished one with Fable 5, and posted the whole process to r/ClaudeAI. A neat before-and-after on how fast the capability gap moves.

Prompting Astra properly. OpenAI's developer team published a guide to how prompts and skills should change for GPT-6 Astra. Worth a read even if you never touch the API; most of it applies to how you talk to ChatGPT.

The sites people are building. Alongside the five-million milestone, OpenAI's Sites post showcased a pixel-art pet game and a browser-based 3D coloring app built entirely from prompts, which says something about what a hosted app builder gets used for.

Two tools worth a look. Neopress builds and grows a website by chatting with it, and Carly spins up agents for scheduling, email and CRM chores. Both are early; try them on something you can afford to lose.

Plan a salary negotiation before the call

Most people walk into a compensation conversation with a number and a hope. This prompt makes ChatGPT or Claude work through your real bargaining position first, then hands you an opening ask, a target, a floor, the objections you will hear and a script. Fill in every bracket honestly; the output is only as good as the position you are really in.

If you have no competing offer, say so; the model will build a case on market data and your achievements instead of a bluff. Advanced tip: paste the employer's likely counter into a follow-up turn and ask it to role-play the hiring manager for three rounds so you can practice out loud.