In San Francisco, it feels like OpenClaw is everywhere. Even, potentially, some places it’s not designed to be. According to posts on social media, people appear to be using the viral AI tool to scrape websites and access information, even when those sites have taken explicit anti-bot measures.
One of the ways they are allegedly doing this is through an open source tool called Scrapling, which is designed to bypass anti-bot systems like Cloudflare Turnstile. While Scrapling, which was built with Python, works with multiple types of AI agents, OpenClaw users appear to be particularly fond of the software. On Monday, viral posts promoting Scrapling as a tool for OpenClaw users started to spread on X. Since its release, Scrapling has been downloaded over 200,000 times.
“No bot detection. No selector maintenance. No Cloudflare nightmares,” reads one viral post this week about the open source tool. “OpenClaw tells Scrapling what to extract. Scrapling handles the stealth.”
Cloudflare is not enthused. The company already blocked previous versions of Scrapling, since users of the open source software kept trying to get around anti-scraping protections. This week, the company was working on a patch for Scrapling’s most recent iteration. “We make changes, and then they make changes,” says Dane Knecht, chief technology officer at Cloudflare. He says the company’s trove of website data and its ability to track trends has given it the upper hand.
“We already had a signal that they’re starting to get a higher ability to get around us,” says Knecht. “The team of security operations engineers had already been working on a new set of mediations.”
Large language models were trained on the corpus of the internet—and the process involved a lot of scraping. In some sense, Scrapling users are following in the footsteps of the original model builders, but on a more individualized scale.
Over the past few years, website owners have attempted to put up additional anti-bot protections, either to block software like Scrapling or to find a way to make money off of the bots trying to access their sites. In turn, Cloudflare has been working overtime to keep blocking increasingly powerful bots attempting to get around these protections.
In July 2024, Cloudflare started to offer its customers additional tools that block AI crawlers, unless the bots pay for access. In less than the span of a year, the company claims to have blocked 416 billion unsolicited scraping attempts.
“I Didn’t Know What I was Getting Into”
As Scrapling gained traction in recent days, crypto enthusiasts capitalized on the attention by launching a $Scrapling memecoin. Karim Shoair, who claims to be the sole developer of Scrapling, posted about the memecoin on X (those posts have since been deleted). After the price skyrocketed for around five hours, $Scrapling quickly fell off a cliff as users sold off their stakes. “Bunch of fucking scammers,” reads one comment on the Pump.Fun site that hosts the coin.
“I didn’t know what I was getting into when people made that coin and I endorsed it,” says Shoair, in a direct message with WIRED. “But once I knew, I didn’t want any association with it and the money I withdrew before will go to charity, I won’t benefit from it in anyway. Or maybe just leave it to be wasted.”
In the fallout of this event, the unofficial GitHub Projects Community account, which has over 300,000 followers on X, deleted its posts from this week highlighting Scrapling’s open source software, and appeared to distance itself from the project. “We do not support, promote, or engage in crypto assets, token offerings, trading activity, or crypto-based fundraising,” it said in a post late Monday night.
Putting the crypto forays aside, most software leaders continue to see agents and autonomous AI tools as the future of the web. Even Knecht from Cloudflare, whose work includes blocking bots from nonconsensual scraping, wants to build toward a world where humans and agents benefit from online data and the wishes of website owners are respected. “I see a path forward for an internet that is both friendly to agents and humans,” he says.
This is an edition of Will Knight’s AI Lab newsletter. Read previous newsletters here.
