Getting your
Trinity Audioplayer ready...I attended this year’s Trust & Safety Professional Association (TSPA) TrustCon conference at the Hyatt Regency in San Francisco, where nearly 1,400 people gathered to discuss how to better manage online risks. Although attendees represented a wide variety of job functions, most work in areas such as moderating online communities, developing and enforcing platform safety policies, combating fraud and abuse and protecting children online. As one would expect, many are now focused on making AI platforms safer and more trustworthy.
Unfortunately, many of the sessions were off limits to the press, but I was able to attend a few, including a fireside chat featuring TSPA Chair Del Harvey, a pioneer who built Twitter’s Trust & Safety organization, and David Orr, head of Safeguards at Anthropic, the company behind the Claude AI assistant.
Old safety concerns, new technology
Harvey argued that AI safety isn’t really a new discipline but rather the next evolution of trust and safety. For years, trust and safety professionals have developed policies and systems to combat spam, abusive content and similar harms. In her view, generative AI is simply the latest, and perhaps most complex, technology requiring those same skills and principles.
The same is true for technology users and their families. As someone who has been writing about online safety since 1994, I can attest that many of the same precautions I’ve recommended over the decades remain just as relevant in the age of AI as they were when people were mostly accessing newsgroups, websites, and later, social media. Critical thinking and media literacy remain essential. Don’t automatically believe everything you read, see or hear and be careful about what you share with AI systems or any online service, because information you enter could ultimately be seen by unintended people or used in ways you didn’t anticipate. And, as always, parents should have occasional conversations with their children about how they are using technology.
Balancing benefits and risks
Orr described Anthropic’s approach as “holding light and shade,” referring to the need to balance AI’s enormous benefits with the risks of misuse, including bioweapons and cyberattacks. He said the company takes a precautionary approach, preferring to identify and address potential risks before releasing new AI models, because retraining them later is both time-consuming and resource intensive. Anthropic also emphasizes “threat modeling,” identifying different types of bad actors, from sophisticated criminals to what he called “chaos actors,” as well as well-intentioned users who inadvertently misuse AI.
In another session, Colton Pond, chief marketing officer of Socure, a risk management company, argued that trust and safety should be viewed not only as a way to reduce risk but as a business advantage. According to research he presented, 97% of business leaders reported at least one growth benefit from investing in trust and safety, including improved customer retention, stronger brand reputation, faster onboarding, lower fraud losses and higher conversion rates.
Pond said trust and safety teams need to rethink their priorities as AI changes the threat landscape. Instead of focusing primarily on content moderation, companies should treat identity as the new security perimeter, defending against bots, fake accounts and falsified identities. They should measure their defenses against today’s AI-driven attacks rather than yesterday’s threats and use technology to support rather than replace human judgment.
False positives
Another key theme was reducing what Pond called the “tax on legitimate customers.” Trust and safety systems should stop bad actors without creating unnecessary friction for honest users. Examples included AI content moderation systems that fail to understand context and fraud detection systems that mistakenly block legitimate customers.
This issue of false positives is also one of my concerns, not just in security but in how we consume information. It’s bad enough that we’re at risk of being fooled by fake images, videos and other misinformation. Equally troubling is how easy it has become to dismiss authentic information as fake.
Someone close to me had to wait an extra six months to receive her Social Security benefits because she assumed an email from the Social Security Administration was spam and hung up on an agency employee, convinced the call was a scam. In reality, the agency simply needed additional information to process her application. More recently, the internet was abuzz over a so-called “proof of life” photograph of Senator Mitch McConnell and his wife, Elaine Chao. I saw numerous social media posts claiming the image was AI-generated or otherwise fake. But according to The Washington Post, which reviewed the original image and its metadata, the photograph was authentic, adding “An independent digital forensics expert also said there appeared to be no evidence that the image is fake.”
Teens use of AI
In another session, Nina Bual, co-founder of Cyberlite, cited research from Common Sense Media, Internet Matters, Cyberlite and Microsoft showing that teens are increasingly turning to AI companions for support, advice and social interaction. According to the research she presented, 72% of U.S. teens have used AI companions, while nearly 40% of U.K. teens rely on chatbots for advice or companionship. Perhaps most surprising, about one-third of students said they worry more about losing their critical thinking skills through overreliance on AI than about AI taking their jobs.
I share those concerns. Critical thinking is more important than ever because both young and older people need to question what AI tells them, at least until these systems become far more reliable. And although cognitive development is an obvious concern for young people, it’s also important for older adults, who may already face age-related memory challenges. As someone who increasingly relies on AI for research, it’s something I think about for myself as well.
Bual said the best way to teach young people about AI safety is to meet them where they are. Her research found that teens overwhelmingly prefer clear data, statistics, and real-world case studies over traditional awareness campaigns. That’s a lesson worth remembering. Whether we’re designing AI systems or simply using them, the goal shouldn’t just be to embrace the technology. It should be to use it wisely, thoughtfully, and with a healthy dose of skepticism.
Larry Magid is a tech journalist and internet safety activist. Contact him at larry@larrymagid.com.