Grok's safety meltdown, EU figures stand behind DSA and red-teaming office snacks

Hello and welcome to the Everything in Moderation's Week in Review. It's written by me, Ben Whitelaw and supported by members like you. Happy new to all EiM subscribers.

2026 may prove me wrong but the first EiM edition of the year has historically been a good barometer of what's come in the subsequent 12 months. In 2025 (EiM #276), it was Meta’s abandonment of T&S, the increasingly important role of safety tech vendors and growing EU ire towards X/Twitter (which we saw play out in December) while in 2024 (EiM #229) it was slow US lawmaking and AI’s emerging T&S issues.

So what does this year’s opener suggest? To me, it's a world where AI systems become the primary cause of user safety; where governments shift from consultation to intervention (not always successfully); and where internet users continue to disagree — often loudly — about whether speech or safety should reign supreme. EiM plans to follow how those tensions show up in law, product decisions, and the everyday work of T&S teams.

I’m still working through EiM’s goals for 2026 and plan to share more in the coming weeks. But if you or your organisation want to partner on project, sponsor a future edition or share a stage in 2026, drop me a line.

T&S Insider, EiM's Monday newsletter helmed by the experienced T&S pro Alice Hunsberger, will be back in your inbox next week. As ever, if you want to tailor your newsletter preferences, you can do so via Your Account. Here's your first roundup of 2026 – BW

SPONSORED BY Tech COALITION, Uniting industry to fight child sexual abuse

Last year, Tech Coalition members deepened industry-wide collaboration to protect children online. Through shared tools, safety innovation, and working groups, members spent nearly 1,000 hours collaborating to tackle persistent and emerging forms of online child sexual exploitation and abuse.

This collective effort is strengthening industry’s ability to detect harm faster, share critical insights, and drive practical solutions at scale.

SEE THE IMPACT

Policies

New and emerging internet policy and online speech regulation

The architects of the EU’s Digital Services Act and Digital Markets Act have written a punchy essay in response to the US’ decision to ban five Europeans, including former commissioner Thierry Breton (EiM #259), from entering the country. The most interesting part of the piece, published by Project Syndicate, is the assertion that Europeans have been “captured by big-tech platforms” which poses “risks that can no longer ignore”. Expect the digital sovereignty verbal sparring— as well as the investment into European startups — to further ramp up this year.

As of January 1st, Malaysia’s own Online Safety Act will impose new duties on the largest social media and internet messaging platforms in the country after the likes of Facebook, Instagram, WhatsApp, X/Twitter and YouTube all failed to obtain the necessary license last year. The Malaysian Communications and Multimedia Commission (MCMC) has subsequently opted-in any platform with more than 8 million users to fulfil obligations including preventing CSAM, financial frauds and scams but also what it calls “obscene / lewd” and “indecent” content. The South China Morning Post has more details.

Also in this section...

• Meet the new tech laws of 2026 (The Verge)
• How to Make Sense of Trump’s TikTok Deal (Tech Policy Press)

Products

Features, functionality and technology shaping online speech

A new study published by Meta’s researchers on Christmas Eve (whose idea was that?!) looks at the promising use of reinforcement learning (RL) — a technique for aligning general-purpose LLMs — for content moderation. The paper gets maths-y very quickly but the gist seems to be that RL a more effective means of than prompting and supervised fine tuning, which are the main ways to refine frontier models for content review right now.

A fun story from the Wall Street Journal, which let Anthropic’s Claude model autonomously run its office vending machine — leading it to lose hundreds of dollars through poor pricing, over-compliance, and being talked into bad deals. Why is it interesting to EiM readers? The experiment was led by Anthropic’s frontier red team and the very good video with WSJ reporter Joanna Stern includes an interview with Logan Graham, head of the team. 

Platforms

Social networks and the application of content guidelines

Grok became the big story this week after it flooded X/Twitter with graphic sexual content on demand — including sexualised descriptions of women and girls — often in ways that would violate X’s own platform rules. Not only that but Wired also found many examples of “overwhelmingly sexual content” hosted and shared via grok.com, suggesting that the problem is much bigger than what is visible publicly on the Elon Musk-owned platform.

Under pressure: The story attracted attention from elected officials in Malaysia, France, and India while UK regulator Ofcom “made urgent contact with X and xAI to understand what steps they have taken to comply with their legal duties to protect users in the UK”. That pressure seems to have done something: Grok this morning limited image generation to paying subscribers.

The parents a 16-year-old who took his own life after being extorted on Instagram, has spoken to The Guardian about their “duty to warm other parents”. Ros and Mark Dowey are part of a lawsuit filed in Delaware by the Social Media Victims Law Center and said “nothing had changed”  since their son Murray died in 2023.

Also in this section...

• How millennials fell out of love with the internet (Vox)
• TikTok Saga Moves Quietly Toward Understated Ending (Bloomberg)

People

Those impacting the future of online safety and moderation

Commenters on news websites are a special breed. I should know; I spent more than a decade negotiating with the most vicious of them when I was a fresh-faced 20-something digital editor. It was that experience — running a team of moderators, setting policy and trying to enforce it — which led me to found Everything in Moderation in 2018.

While I don’t get to hang out below the line anymore— shout out “Baron von Baldness” or “Proverbial” and “DrJ” — I did enjoy the reader comments (and emails) on The Guardian’s call out for answers to the question “Should we turn the internet off?”.

Some are predictably quaint (“Bring back Argos catalogues”), some are sanguine (“There are so many other areas where the spirit of the early internet lives on.”) and some are silly (“Thanks to Google maps, I am walking in circles in the middle of Epping Forest!!”).

Naturally, few of them recognise that they wouldn’t be able to share their views on the Guardian without it. But, then again, that’s the beauty of commenters.

Posts of note (new jobs edition)

Handpicked posts that caught my eye this week

• "Are you thinking "new year, new job"? At Resolver Trust & Safety we have two open roles" - Henry Adams is looking for a Kurdish and Urdu speakers to join the team.
• "I’m hiring for a Principal of Trust & Safety GenAI Strategy to lead the safety strategy for Google’s agentic features." - One helluva role from Christina Heggie .
• "Happy New Year! 🎉 TikTok Trust and Safety is still HIRING in 2026!" - Marco Galvan has ML engineering, ops roles and red teaming gigs up for grabs.