
What happened
Russian publisher Eksmo-AST trained an AI agent to flag books and passages risking violation of Russia's post-2022 censorship laws, ran about 20,000 checks, and handed 600 volumes to external experts, its president Oleg Novikov told state TV.
Why it matters
The publisher says it is "on the frontier of technological progress" — meaning a domestic firm has applied AI to compliance with the restrictive criteria rather than to publishing itself.
What to watch
Whether the flagged 600 volumes are ultimately removed from shelves depends on the external experts' reviews, so the AI agent's role stays a first-pass filter for now.
WHO IT HITSPublishers and media compliance teams operating under restrictive speech laws are the clearest audience — this shows an AI agent can be used as a first-pass filter for legally risky content at scale, with human experts still making the final call.
Ask the AI about this article →
Summaries like this, in your inbox every morning.
Eksmo-AST's move shows how quickly AI tooling can be repurposed when the operating environment changes abruptly. Its trigger was a legal one: Russia's censorship laws, imposed after the 2022 invasion of Ukraine, created new restrictive criteria that the publisher had to apply retroactively to a catalogue built over two decades. Training an AI agent to scan that backlist for risky books and passages was the company's answer, and the roughly 20,000 checks it ran suggest the scale of the job.
The detail that matters most is where the AI's authority stops. Of the flagged material, 600 volumes were sent to external experts for additional review — a second, human layer before any decision. So the agent is not the final judge; it is a first-pass filter that narrows what people then examine.
Oleg Novikov's framing — "We're here on the frontier of technological progress" — reads as an attempt to present the tooling as a capability story rather than a censorship story. How the 600 volumes fare in external review, and whether the AI agent becomes a standard part of the publisher's workflow, is the test of whether this stays a compliance measure or becomes a permanent publishing tool.
For example, today's edition would include:
AI-summarized, only the topics you pick — one digest a day via Email, Slack, or Discord.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. Q&As are published on this page for other readers too.
Anthropic CEO Dario Amodei published a 3,800-word essay, "We Must Pace the Frontier," urging policymakers to a…

Polling for Alex Bores' new group Who Decides found half of 1,502 registered voters say AI is moving too fast…

On a visit to Hamilton County, Tennessee, the author found ChatGPT blocked on school wifi but reached it insta…

At the Yale CEO Caucus, 93% of attendees disagreed with Trump's Truth Social post calling AI warnings a hoax…

Anthropic CEO Dario Amodei urged the US and China to cooperate on slowing AI in his Saturday blog post "We Mus…

Meta announced on Sept. 15 the global rollout of Meta One, a subscription combining premium features across it…
