Facebook moderation leak, UK Online Safety Act and Reddit's Rules Hub show the same fight

The leak landed at an awkward moment, in the middle of the UK general election campaign. The Conservatives were pushing for fines on sites that fail to remove illegal content quickly, Ars Technica reported on 22 May 2017.

Facebook's defence, then and largely now, is scale and legal category. The leaked documents show the company drawing a line at content that could credibly cause real world harm, and requiring certain details before it treats a threat as credible. Under those rules, violent misogynistic posts and calls to beat up fat kids stayed up. Threatening a president did not, partly because it is a criminal act in the US. The company has long insisted it is a platform, not a publisher. In Europe it can lean on the E-commerce Directive, whose Article 15 says hosting services are not obliged to monitor what they transmit or store. That position gets harder to hold every time Facebook intervenes in content, and a moderation rulebook is exactly that. Facebook's global policy management boss, Monika Bickert, told The Guardian that a "diverse global community" makes consistent line drawing impossible, and that the gap between satire and inappropriate content is sometimes very grey.

Six years later, the UK stopped asking nicely.

The Online Safety Bill received royal assent on 26 October 2023 and became the Online Safety Act, The Verge reported. It covers underage access to pornography, anonymous trolls, scam ads, nonconsensual intimate deepfakes, child sexual abuse material and terrorism content. Ofcom enforces it in three phases.

"Our new powers are not about taking content down," Ofcom chief executive Melanie Dawes said in a statement. "Our job is to tackle the root causes of harm."

Penalties reach £18 million, around $22 million, or 10 percent of global annual turnover, whichever is higher, and bosses can go to prison. WhatsApp and Signal objected to a clause letting Ofcom ask companies to identify child sexual abuse content whether communicated publicly or privately, saying it undermines end-to-end encryption; some providers suggested they would rather leave the UK. The Wikimedia Foundation warned that strict child protection duties sit badly with a service that collects minimal user data, including ages.

Automation moves in, and the rules get fuzzier

The newest turn is that platforms are handing judgement to language models. Reddit is rolling out a moderation suite called Rules Hub, which uses LLMs to evaluate whether a post or comment matches the intent of a rule, The Verge reported on 6 August 2026. Rules Hub has been tested with moderators from over 700 communities. It is expanding to all newly created communities and will be widely available later in 2026.

Reddit says Rules Hub can eventually replace many enforcement workflows communities rely on from Automoderator, which depends on exact keyword and pattern matching. CEO Steve Huffman called Automod hard to learn, hard to maintain and heavily dependent on brittle keyword matching and inherited configurations. The company is also requiring new third-party apps to use its developer platform instead of the API, and says it is exploring changes to old Reddit, with no set timeline.

Set that against Valve. PC Gamer reported on 2 June 2026 that, according to a Bloomberg account citing an anonymous former employee, Gabe Newell snapped at general counsel Karl Quackenbush, who was arguing for more hands-on moderation as Valve weighed embracing outright pornographic content: "What the fuck do I pay you for if that's your opinion?" Steam opened the floodgates anyway, later delisting adult content under pressure from payment processors.

So the same argument keeps recurring in different costumes. Facebook wrote narrow rules and called it neutrality. The UK legislated and let a regulator write the detail. Reddit is replacing brittle rules with models that judge intent. Valve let the market, and occasionally a payment processor, decide. None of them has settled the question the 2017 leak posed: who draws the line, and who is answerable when it moves.