Social networking startup Bluesky, which is building a decentralized alternative to X (formerly Twitter), offered an update on Wednesday about how it’s approaching various trust and safety concerns on its platform. The company is in various stages of developing and piloting a range of initiatives focused on dealing with bad actors, harassment, spam, fake accounts, video safety, and more.
To address malicious users or those who harass others, Bluesky says it’s developing new tooling that will be able to detect when multiple new accounts are spun up and managed by the same person. This could help to cut down on harassment, where a bad actor creates several different personas to target their victims.
Another new experiment will help to detect “rude” replies and surface them to server moderators. Similar to Mastodon, Bluesky will support a network where self-hosters and other developers can run their own servers that connect with Bluesky’s server and others on the network. This federation capability is still in early access. However, further down the road, server moderators will be able to decide how they want to take action on those who post rude replies. Bluesky, meanwhile, will eventually reduce these replies’ visibility in its app. Repeated rude labels on content will also lead to account-level labels and suspensions, it says.
To cut down on the use of lists to harass others, Bluesky will remove individual users from a list if they block the list’s creator. Similar functionality was also recently rolled out to Starter Packs, which are a type of sharable list that can help new users find people to follow on the platform (check out the TechCrunch Starter Pack).
Bluesky will also scan for lists with abusive names or descriptions to cut down on people’s ability to harass others by adding them to a public list with a toxic or abusive name or description. Those who violate Bluesky’s Community Guidelines will be hidden in the app until the list owner makes changes to comply with Bluesky’s rules. Users who continue to create abusive lists will also have further action taken against them, though the company didn’t offer details, adding that lists are still an area of active discussion and development.
In the months ahead, Bluesky will also shift to handling moderation reports through its app using notifications, instead of relying on email reports.
To fight spam and other fake accounts, Bluesky is launching a pilot that will attempt to automatically detect when an account is fake, scamming, or spamming users. Paired with moderation, the goal is to be able to take action on accounts within “seconds of receiving a report,” the company said.
One of the more interesting developments involves how Bluesky will comply with local laws while still allowing for free speech. It will use geography-specific labels allowing it to hide a piece of content for users in a particular area to comply with the law.
“This allows Bluesky’s moderation service to maintain flexibility in creating a space for free expression, while also ensuring legal compliance so that Bluesky may continue to operate as a service in those geographies,” the company shared in a blog post. “This feature will be introduced on a country-by-country basis, and we will aim to inform users about the source of legal requests whenever legally possible.”
To address potential trust and safety issues with video, which was recently added, the team is adding features like being able to turn off autoplay for videos, making sure video is labeled, and ensuring that videos can be reported. It’s still evaluating what else may need to be added, something that will be prioritized based on user feedback.
When it comes to abuse, the company says that its overall framework is “asking how often something happens vs how harmful it is.” The company focuses on addressing high-harm and high-frequency issues while also “tracking edge cases that could result in serious harm to a few users.” The latter, though only affecting a small number of people, causes enough “continual harm” that Bluesky will take action to prevent the abuse, it claims.
User concerns can be raised via reports, emails, and mentions to the @safety.bsky.app account.
Keep reading the article on Tech Crunch
LinkedIn may have trained AI models on user data without updating its terms.
LinkedIn users in the US — but not the EU, EEA, or Switzerland, likely due to those regions’ data privacy rules — have an opt-out toggle in their settings screen disclosing that LinkedIn scrapes personal data to train “content creation AI models.” The toggle isn’t new. But, as first reported by 404 Media, LinkedIn initially didn’t refresh its privacy policy to reflect the data use.
The terms of service have now been updated, but ordinarily that occurs well before a big change like using user data for a new purpose like this. The idea is it gives users an option to make account changes or leave the platform if they don’t like the changes. Not this time, it seems.
So what models is LinkedIn training? Its own, the company says in a Q&A, including models for writing suggestions and post recommendations. But LinkedIn also says that generative AI models on its platform may be trained by “another provider,” like its corporate parent Microsoft.
“As with most features on LinkedIn, when you engage with our platform we collect and use (or process) data about your use of the platform, including personal data,” the Q&A reads. “This could include your use of the generative AI (AI models used to create content) or other AI features, your posts and articles, how frequently you use LinkedIn, your language preference, and any feedback you may have provided to our teams. We use this data, consistent with our privacy policy, to improve or develop the LinkedIn services.”
LinkedIn previously told TechCrunch that it uses “privacy enhancing techniques, including redacting and removing information, to limit the personal information contained in datasets used for generative AI training.”
To opt out of LinkedIn’s data scraping, head to the “Data Privacy” section of the LinkedIn settings menu on desktop, click “Data for Generative AI improvement,” and then toggle off the “Use my data for training content creation AI models” option. You can also attempt to opt out more comprehensively via this form, but LinkedIn notes that any opt-out won’t affect training that’s already taken place.
The nonprofit Open Rights Group (ORG) has called on the Information Commissioner’s Office (ICO), the U.K.’s independent regulator for data protection rights, to investigate LinkedIn and other social networks that train on user data by default. Earlier this week, Meta announced that it was resuming plans to scrape user data for AI training after working with the ICO to make the opt-out process simpler.
“LinkedIn is the latest social media company found to be processing our data without asking for consent,” Mariano delli Santi, ORG’s legal and policy officer, said in a statement. “The opt-out model proves once again to be wholly inadequate to protect our rights: the public cannot be expected to monitor and chase every single online company that decides to use our data to train AI. Opt-in consent isn’t only legally mandated, but a common-sense requirement.”
Ireland’s Data Protection Commission (DPC), the supervisory authority responsible for monitoring compliance with the GDPR, the EU’s overarching privacy framework, told TechCrunch that LinkedIn informed it last week that clarifications to its global privacy policy would be issued today.
“LinkedIn advised us that the policy would include an opt-out setting for its members who did not want their data used for training content generating AI models,” a spokesperson for the DPC said. “This opt-out is not available to EU/EEA members as LinkedIn is not currently using EU/EEA member data to train or fine tune these models.”
TechCrunch has reached out to LinkedIn for comment. We’ll update this piece if we hear back.
The demand for more data to train generative AI models has led a growing number of platforms to repurpose or otherwise reuse their vast troves of user-generated content. Some have even moved to monetize this content — Tumblr owner Automattic, Photobucket, Reddit, and Stack Overflow are among the networks licensing data to AI model developers.
Not all of them have made it easy to opt out. When Stack Overflow announced that it would begin licensing content, several users deleted their posts in protest — only to see those posts restored and their accounts suspended.
Keep reading the article on Tech Crunch