The slop storm is thick and blinding. As a young, loopy David famously asked after his dentist appointment: “Is this going to be forever?” We’d love to say no, but we have to be honest. For now, all we can do is fashion a countermeasure.
At stake is our shared understanding of reality. A vision for how to proceed. To find our bearings, we will create a lens. Slop-vision goggles.
What is this? This is us getting the goggles into focus.
Bear with us; it’s Glitchy.
North Korea is minting meme coins to launder the Binance hack. You read that right, and it’s the exact headline of a recent post from Elliptic, which explains how the hackers have passed hundreds of thousands of dollars in real, stolen value by creating their own memecoins on the Tron, Solana, and NEAR blockchains. Here’s how they are doing it, according to the analytics firm:
Create a token. The launderers create a new token using one of the public token-launch tools available on (a given) blockchain.
Open a pool. They pair the new token with a real asset, such as NEAR or a stablecoin, in a liquidity pool on a decentralised exchange, where anyone can swap one for the other.
Buy with stolen funds. One set of wallets pays stolen funds into the pool and receives the new token in return.
Sell and collect. A second set of wallets sells the token back into the same pool and withdraws the real asset.
The stolen money never moves from a buying wallet to a collecting wallet, Elliptic explains. “A wallet whose history shows trades in a new meme coin looks like a speculator rather than a launderer, unless someone checks where its money came from.” And on the other side: “To a casual observer, the collecting wallets look like ordinary traders who have sold a token at a profit.”
Tron, Tether, and Chainalysis have formed a new crypto-crime-fighting organization. The Digital Crime Task Force will target illicit activity involving Tether’s USDT stablecoin on the Tron blockchain. The organization ostensibly replaces one that had included Tron, Tether, and Chainalysis rival TRM Labs. That organization, which was called the T3 Financial Crime Unit, had quietly disbanded in August, according to recent reporting from Bloomberg. According to Decrypt, the new group also aims for a fresh approach:
The pitch is a shift from reactive to proactive: rather than waiting for a victim to report a crime or an investigator to flag tainted funds, the group says it will proactively trace criminal networks and coordinate asset freezes, drawing on Chainalysis‘s intelligence and AI-powered investigative tools. It will also accept tips and freeze requests from police.
Google DeepMind researchers used AI to formally verify certain of Signal’s “key cryptographic guarantees.” That’s according to Wired, which notes that besides reassuring trust in the messaging app’s cryptography, the research sheds light on the power of AI-aided formal verification, a process that uses math to determine whether a piece of software doesn’t contain critical vulnerabilities. That said, as Wired’s report also flags, at least one key detail has not been publicized:
The open question across the tech sector, though, is whether AI-automated formal verification can truly scale—and whether it can do so at a workable cost. DeepMind conducted the Signal analysis free of charge as a research project, and Google declined to share the cost of the analysis. The company also declined to comment on whether it has plans to release a commercial formal verification product or service.
Meta is taking its “AI age-estimation technology” global. For several years, the firm has been developing technology that uses AI “to find people we suspect to be teens, even if they use an adult birthday.” In May, it started expanding how it uses the tool, rolling it out for Instagram in the EU and Brazil and Facebook in the US. Now the company says the technology will be used globally on Instagram, Facebook, and Threads. How does it work? The announcement in May described “using AI technology to analyze entire profiles for contextual clues—such as birthday celebrations and mentions of school grades—to determine if an account likely belongs to someone under age.”
There’s also “visual analysis”:
This technology allows our AI to scan photos and videos for visual clues about a person’s age, giving us an additional signal, especially on accounts where text-based indicators may not be present. We want to be clear: this is not facial recognition. To help determine if a person is underage, our AI looks at visual cues in a photo or video. This is similar to how a person might spot visual cues in a photo that suggest someone is underage. Our AI does not identify the specific person in the image.
In July, Meta updated the May post to “clarify that our AI does not use biometric data to identify whether a person is underage,” adding that it also “does not create facial fingerprints or identify who is in the photo.”
Wired went down the “abliteration” rabbit hole. It found cypherpunks. Not sure about you, but our timelines are full of folks waxing this way and that about abliterated LLMs. At a high level, it means the guardrails built into AI models that prevent them from responding to certain requests have been removed. A recent Wired deep dive includes some helpful technical context:
The term combines ablation, meaning a surgical removal, with obliteration. It emerged after a group of academic researchers showed, in 2024, that a model’s refusal to comply with a request has a remarkably simple mechanism. There is effectively a master switch in a model’s weights that determines whether it will say no. Removing this feature, they found, stops the model from ever refusing to answer a prompt.
Other researchers quickly picked up on this technique, tweaked it, and developed a number of tools for abliterating open-weight models.
Naturally, people disagree about whether this capability is good. Then again, there’s no disagreement that it could help people cause harm. Perhaps most importantly, as the Wired piece notes, there’s probably no stopping abliteration without banning open-weight models. Using policy to snuff out open-source software tools is unlikely to work; that’s part of the point! Besides, abliteration advocates argue, tech companies and governments shouldn’t be the only ones with the power to determine the guardrails. (Sound familiar, crypto folks?)
Wired’s reporter noticed something, too:
The more I learned about abliteration and the people involved in it, the more I came to see them as the direct descendants of an earlier techno-libertarianism movement, the cypherpunks. Those engineers, lawyers, and campaigners fought to prevent government restrictions on cryptography at a time when many feared it would be misused by terrorists and child pornographers. The cypherpunks played a role in making public-key cryptography a cornerstone of internet communications and commerce.
Agents are starting to buy things with credit cards. Banks aren’t loving it. Like it or not, the future seems near at hand in which AI makes purchases on our behalf. Modern-day payment networks are messy and full of middlemen, but the lion’s share of risk is usually borne by banks or credit cards. They naturally have strong opinions about some of the weirdness they’re seeing in the emerging world of AI commerce. From The Wall Street Journal:
Some payment executives say they are already seeing higher returns from agentic shopping compared with when consumers make their own purchases, as well as elevated fraud cases, such as from agents purchasing items from fake online stores that never deliver the products. And there are concerns about AI supercharging points gaming, which could make premium cards more expensive for banks.
Some companies are banning agents and agentic payments. Others are embracing it. The battle is on, and the bots are winning.


