Miasma is a lightweight Rust server with a specific purpose: trap web scrapers in a loop of procedurally generated garbage, keeping them busy consuming bandwidth and CPU while feeding low-quality data into whatever training pipeline sent them. The scraper gets trapped in a maze of hidden links invisible to humans but visible to bots, leading down rabbit trails of junk content.
The technical implementation is clean enough. It has a minimal memory footprint, linear scaling with concurrent connections, configurable link density and gzip compression to manage egress costs. The architecture is honest about its limits - it returns 429s when capacity is exceeded rather than queuing indefinitely. The README even specifies how to allow "friendly" bots and scrapers. As protest infrastructure goes, it is well-built.
That word - protest - is worth sitting with.
The instinct behind Miasma is legible and not unreasonable: large commercial actors are harvesting content at scale without compensation or permission, and the people producing that content have limited recourse should they want it. Doing something feels better than doing nothing, and Miasma offers something to do.
But the economics of the approach are worth examining. The mechanism works by consuming your resources - bandwidth, CPU, egress budget - in proportion to attacker volume. (If you only have so much outgoing bandwidth, then Miasma's going to help you use it up.) The scraper's marginal cost is low because they're already operating at scale. The defender's cost scales up with engagement. It also depends on scrapers following embedded links, which is precisely the behavior that modern quality-filtering pipelines are designed to suppress. The operators most likely to get caught in the loop are smaller, less sophisticated actors. The large commercial targets are largely elsewhere.
Consuming more of their resources hurts them, yes, but those resources are already being consumed at scale - and adding more consumption is likely to harm the communities that host the large data centers more measurably than the AIs themselves.
None of this means Miasma is wrong to exist. It means the tool solves a different problem than the ones that typically make people angry.
The countermeasures that have materially changed the landscape - AP's licensing negotiations, the New York Times lawsuit, class action suits against Anthropic and others, ongoing legislative efforts around training data consent - operate on incentive structures. They change what it costs to scrape, not just whether the scraped data is clean. Technical countermeasures are, at best, friction. Legal and economic pressure is where the actual leverage lives.
Miasma is worth knowing about. The Rust implementation is a reasonable read for anyone interested in this corner of the problem space.