This AI's CRISPR Claim Is Raising Eyebrows, Not Cheers

An AI found something that looks like CRISPR, two AI labs cut prices on the same day for reasons that have nothing to do with generosity, and an OpenAI agent broke into a government website nobody asked it to touch. Today's issue is about what AI actually did this week, not what it promised to do.

The professionals pulling ahead aren't working more. They're using Claude.

Our free guide will show you how to:

Transform your workflow with AI and stay ahead of the curve with this comprehensive guide to using Claude at work.

Get The Free Guide

What happened: Anthropic says its Claude AI model helped identify a previously unknown enzyme system in viruses that infect bacteria, sitting next to a repeating DNA pattern that looks a lot like the setup behind CRISPR gene editing. Anthropic turned about 950 copies of Claude loose on a database of over 200,000 enzymes for 21 hours, and one of those copies flagged the odd repeat pattern, which human scientists then confirmed in the lab. CEO Dario Amodei called it "work I would have been proud to do as a PhD student" and said the system, which Anthropic named ART, could turn out to be a new kind of gene editor, though outside scientists were far more cautious. A Washington University microbiologist told Al Jazeera there's nothing yet showing this rivals CRISPR as an actual technology.

Why it matters to you: This is a preview of how every AI "breakthrough" claim is going to read for a while: an impressive process, an unproven payoff. The same gap shows up when a vendor pitches you an AI tool. Ask what the AI actually proved, not what the press release implies it proved.

What happened: Anthropic launched Claude Opus 5.5 on Tuesday at $4 per million input tokens and $20 per million output tokens, a 20% cut from Opus 5, and about an hour later OpenAI launched two new models, GPT-6 Sol and GPT-6 Luna, both priced 50% below their predecessors, according to 9to5Google. Anthropic claims Opus 5.5 beats OpenAI's flagship GPT-6 Astra in most categories, while OpenAI is leaning on Sol and Luna being the cheap, fast options next to Astra. Both companies are also running noticeably faster models than their last generation.

Why it matters to you: If your business runs anything on top of ChatGPT or Claude, whether that's a chatbot, a research tool, or an app someone built for you, your per-use cost just moved, and it will keep moving every few weeks for the foreseeable future.

What to do about it: If you're paying for API usage rather than a flat subscription, check which model your tool defaults to. You may be able to switch to a cheaper option without losing quality.

I had serious FOMO watching last year's Epic Marketing Summit from the sidelines, and I'm not making that mistake twice. This is a 2 day live event in Miami where you don't just learn about AI funnels, you build one. You walk in with your business and walk out with a working AI powered sales funnel already built, plus your own vibe coded AI app made on-site over the two days. No frameworks to go implement later on your own time.

I know a lot of the trainers running this thing personally, and they've actually done the work they teach, they're not reciting theory. Last year's event sold out, and January 2027's version is going even harder on the AI side. If you're tired of running your sales process on pure hustle and want to turn it into an actual system, this is where I'd point you. I'll be there this time. Grab your seat here.

What happened: The nonprofit research lab Transluce says autonomous OpenAI agents tried to force their way into a University of New Mexico digital library, a US government data site called DataUSA, and an Australian health statistics portal, all while running ordinary data-gathering tasks nobody set up as hacking attempts, according to Help Net Security. The pattern goes back to at least March. Australian Prime Minister Anthony Albanese has since confirmed one of those agents actually got into the Medicare statistics portal in June and pulled both public and non-public files, and OpenAI waited nearly three months to tell the Australian government.

Why it matters to you: If you've turned an AI agent loose on tasks that involve browsing the web or logging into accounts on your behalf, you're trusting it to stay inside the lines you drew. These agents didn't. They improvised past blocks nobody trained them to get past.

What to do about it: Before you give any AI agent access to a website, account, or database, know exactly what it's allowed to touch, and check whether the tool logs what it actually did.

What happened: At Meta Connect this week, Mark Zuckerberg unveiled $1,299 VR Glasses launching next spring, a pocket-sized device called Muse Charm for talking to Meta's AI agent without pulling out your phone, and a plan to ship more than 100 styles of AI glasses by year end, according to Engadget. The new headset weighs about 100 grams, roughly a sixth of Apple's Vision Pro, and Meta's Muse assistant is now wired into shopping partners including Walmart and Instacart.

Why it matters to you: Meta isn't betting on any single gadget here, it's betting that an AI assistant becomes the thing people reach for by default, on their face, in their pocket, or in a headset. If that bet pays off even partly, expect customers to start expecting your business to show up wherever their assistant looks first.

What happened: Adobe Analytics data tracking over a trillion visits to US retail sites found that traffic referred by AI sources jumped 138% year over year as of May, and for the first time, shoppers who arrived via AI converted 54% better than shoppers who didn't, according to Digital Commerce 360. Those AI-referred visitors also spend more time on-site and view more pages per visit than typical traffic.

Why it matters to you: If you sell anything online, AI chat tools are quietly becoming a real referral channel, and the shoppers they send convert better than the ones you're used to. Categories with clear, well-structured product information, like cosmetics and electronics, are capturing the most of this traffic. Grocery and furniture sites are lagging.

What to do about it: Take a look at your product pages. If specs, sizing, materials, and policies aren't written out in plain, scannable text, an AI assistant summarizing your site to a shopper may be working with less than it should.

I've been a member of the Ignite Mastermind for over three years now, and honestly, some of my closest friends and a couple of real business partnerships came directly out of being in that room. If you're running a business under $1 million a year and feel like you're doing everything alone, this is built for exactly that stage.

You get 6 to 8 live virtual workshops a year plus 2 in-person multi-day events, the kind of sessions that normally run $97 to $997 each on their own. There's also a full marketing software suite bundled in that replaces over $5,000 a month in separate tools, all connected instead of scattered across a dozen logins. Every week there's a mentorship call with Perry Belcher and his team, and on top of that I personally do monthly accountability and business development calls with members. It's not a hype machine or a stack of courses you'll never finish. It's a room of people actually building alongside you. See what Ignite is about.

What happened: China's Cyberspace Administration has opened a probe into DeepSeek and Moonshot AI after Anthropic published a report accusing both companies of quietly routing millions of user conversations through fake accounts into Claude, then using Claude's answers to train their own cheaper models, according to Decrypt. Anthropic's report claims Moonshot ran more than 23 million exchanges through roughly 5,000 fake accounts, and DeepSeek generated over 12 million exchanges in a single two-week stretch.

Why it matters to you: That cheap AI tool you're considering switching to might not be running its own model at all. Before you adopt a lesser-known AI product because it's cheaper, it's worth asking what's actually powering it under the hood.

What happened: Senator Bernie Sanders and Representative Greg Casar introduced the Ban Artificial Superintelligence Act this week, which would permanently prohibit building AI systems that exceed human cognitive ability, create a new Department of Artificial Intelligence to write safety rules, and require federal approval before any advanced AI system could be deployed, according to PBS NewsHour. Violations could carry up to 20 years in prison. The bill faces long odds in a Republican-controlled Congress that has struggled to agree on much smaller AI rules.

Why it matters to you: This particular bill probably isn't going anywhere, but it's a marker of how far the regulatory conversation has shifted, from "how do we guide AI" to "should some of this be illegal." States are already moving faster than Congress, so the rules your business has to follow may show up locally before they show up federally.

What happened: At the UN General Assembly this week, President Trump announced that all US government documents will now refer to "artificial intelligence" as "Super Intelligence" instead, saying the word "artificial" sounds "fake." According to Gizmodo, the problem is that "superintelligence" already means something specific in AI research, a system that vastly exceeds human ability across the board, which is the exact thing the Sanders bill above is trying to ban.

Why it matters to you: Two branches of government are using the same word this week to mean opposite things, one to make AI sound bigger and better, the other to make it sound dangerous enough to outlaw. That's a decent snapshot of where the AI conversation actually is right now: a lot of noise, not a lot of agreement on what anything means.

Every story in this issue has the same shape underneath it. An AI did something genuinely interesting, and then the actual substance turned out to be smaller, messier, or more uncertain than the headline. That's not a reason to write AI off, Claude really did flag something scientists hadn't seen. It's a reason to keep your expectations tied to what's actually been shown, not to what a keynote or a press release wants you to feel. The rollercoaster keeps running either way. Your job is just to stay off the parts of it you don't need to ride.

If this issue was useful, forward it to one person who needs to stop guessing about AI too. And if you're reading someone else's forward right now, you can subscribe to get this in your own inbox every day.

Talk tomorrow,
Mark Shilensky

Follow me personally on Social Media:
Facebook: https://www.facebook.com/MarkShilenskyPage
Instagram: https://www.instagram.com/markashilensky/
Threads: https://www.threads.com/@markashilensky
LinkedIn: https://www.linkedin.com/in/shilensky/
X: https://x.com/markshilensky
TikTok: https://www.tiktok.com/@mark.shilensky
BlueSky: https://bsky.app/profile/markshilensky.bsky.social