AI Briefing: Ad tech world debates how to address CSAM concerns

Every Adalytics report sparks debate around whether ad tech vendors and brand safety partners are properly managing advertiser dollars. But the latest? It hit a nerve.

After Adalytics revealed ads from major brands were served alongside explicit and harmful content, attention from Congress has intensified the debates. Reactions range from especially loud critiques to especially noticeable silence since U.S. Senators Marsha Blackburn and Richard Blumenthal sent letters this month to the CEOs of Google, Amazon, DoubleVerify, Integral Ad Science, the Media Ratings Council, and Trustworthy Accountability Group (TAG). Many in the industry have condemned weak transparency and accountability, while others point to systemic issues like misaligned incentives and self-regulation failures.

Related Insights

Brand Safety

Amazon, Google and verification vendors among ad tech cohort under fire from U.S. senators over child safety shortcomings Read More

The latest Adalytics report is an “‘emperor has no clothes’ moment for the big brand safety vendors,” Jeromy Sonne and founder of Daypart.ai, told Digiday. Without URL-level transparency, he thinks brands are left vulnerable and overreliant on trusting brand safety partners.

“Instead of pointing fingers, it may make sense to talk about the misalignment of incentives that drives the industry towards an unending and unsustainable quest for efficiency,” IAB CEO David Cohen wrote in one LinkedIn debate. “It is all of our jobs to move the conversation from efficiency at all costs to effectiveness, and starve bad actors from monetization.”

In response to the Adalytics report, Check My Ads Institute (CMAI) filed a formal complaint with TAG, which questions TAG’s transparency, enforcement, and whether its certification serves as a “pay-to-play” system rather than ensuring true accountability. CMAI’s detailed complaint also calls for TAG to conduct a public investigation and to update its certification guidelines & methodology.

While TAG and MRC haven’t issued any sort of formal response to the letters, TAG CEO Mike Zaneis has been actively debating some industry insiders on LinkedIn, including a debate with prominent industry consultant Lou Paskalis. In a Substack post about the issue, Paskalis wrote that addressing CSAM isn’t a nuanced or complex issue, but instead an “all hands on deck issue that demands everyone raise their voices.”

“I don’t care — and neither should you — if it’s less than a thousand ad impressions or less than $5 in contribution wrong is wrong and this is as wrong as it gets,” Pascalis wrote. “You really have to question the judgement of anyone who says otherwise.”

Beyond the current debates, some companies have already taken direct and/or indirect action with new features:

DoubleVerify

• DoubleVerify launched a new “highly illicit” content category, using three years of data from the National Center for Missing & Exploited Children, the same group lawmakers recently questioned companies about. DV also introduced a “P2P sharing and streaming” category to help advertisers avoid peer-to-peer platforms that may host illegal content. It said it has reached out to the FBI and child safety organizations for collaboration and plans to release further research on high-risk content.

Google

• The biggest company in online advertising quietly updated its Transparency Report website to include information about efforts to combat online sexual abuse. The page now outlines Google’s efforts to detect, remove, and report CSAM using AI and hash matching. It also explains how Google removes flagged content, reports it to NCMEC, de-indexes harmful URLs, and tracks law enforcement data to enhance online safety. 
• Google also announced separate tools for built-in protections for kids, including the new use of machine learning to estimate the ages of U.S. users “so that we can apply protections to help provide more age-appropriate experiences.”

Integral Ad Science

• IAS addressed CSAM concerns in a Friday night blog post detailing its review of problematic materials and updates for additional brand safety tools for clients.
• A day earlier, IAS announced separate brand safety features for block-list optimization through a new partnership with Meta, which also announced expanded third-party content blocklists that also include DoubleVerify and Zefr as partners. In its blog post about the Meta updates, stated it uses “industry-leading AI-driven multimedia technology” that includes using AI to analyze “frame-by-frame level” analysis with image, audio and text signals accurate classification.

New questions about the use of AI

There are new questions that come with more reliance on AI for brand safety, especially after lawmakers specifically asked companies why AI systems couldn’t detect and block harmful content. If AI systems aren’t accurately catching the low-hanging fruit — as some experts have argued — what makes the latest updates more likely to detect complex multimedia? Also, if the systems are indeed working like companies say, shouldn’t they be more transparent about how the systems work?

Updates from AI copyright court

The publishing industry had a big win last week in the battle over AI and IP protections. A U.S. judge ruled largely in favor of Thomson Reuters in its lawsuit against Ross Intelligence, rejecting the AI legal startup’s fair use defense. According to the court’s partial order, Ross’s use of Reuters content to create legal research was for commercial use and could impact Reuters’ business — two key criteria for any fair use case. However, the case still has a pending jury trial to address other infringement claims.

In other legal news, a new group of major publishers filed a lawsuit against the Canada-based AI startup Cohere, which provides LLMs to enterprise companies. The lawsuit, filed by the News Media Alliance, lists plaintiffs including The Atlantic, Forbes, the Guardian, Politico, Vox Media & various newspapers. Until now, Cohere has escaped becoming a defendant in publishers’ AI-related copyright lawsuits. (Unlike rivals like OpenAI, Microsoft and others facing a rising number of legal battles.)

BBC releases new study about AI accuracy

The BBC released new research that raised new doubts about the ability of AI assistants to accurately answer questions about news. The study, conducted over the course of a month, analyzed answers from chatbots including ChatGPT, Perplexity, Microsoft Copilot and Google Gemini. In an article about the report, BBC CEO Deborah Turness said BBC News is “already forging ahead” with AI tools for journalism, but cautioned against the risks that come with AI-distorted content.

“The price of AI’s extraordinary benefits must not be a world where people searching for answers are served distorted, defective content that presents itself as fact,” Turness wrote. “In what can feel like a chaotic world, it surely cannot be right that consumers seeking clarity are met with yet more confusion.

The findings could have implications for other publishers that have AI content deals with OpenAI, Perplexity and others. Here’s a look at some stats from the report:

• 51% of AI-generated answers about news topics had “significant issues.”
• 91% of responses had “some issues”
• 19% of answers citing BBC content had incorrect statements, numbers and dates
• 13% of quotes from BBC articles were either altered or didn’t appear anywhere in the cited story.

Prompts and Products – Other AI news and announcements

• Last week’s AI Action Summit in Paris brought together representatives from over 100 countries to address public service AI, future of work, innovation and culture, AI trust, and global governance. The event also included the launch of a $400 million public interest AI fund called Current AI.
• Appier, a Taiwanese marketing tech company, announced a $38.7 deal to buy AdCreative.AI, a Paris-based startup focused on AI-driven ad creation and optimization. 
• Agence France-Presse (AFP) and Nikon said they successfully tested a “proof of concept” to verify the authenticity of whether images have been altered by AI. 
• Adobe released its new Firefly Video model, which features new ways to create AI-generated videos using text prompts, reference images and other tools.

AI-related stories from across Digiday:

• Rundown: FTC’s potential priorities for ad tech, data and AI
• How Media By Mother’s CEO wrote a quick, noteworthy pitch using generative AI 
• CourtAvenue’s latest acquisition dives into the world of bots — the less evil kind
• The Trade Desk missed its revenue target for the first time, here’s how CEO Jeff Green pledged to fix it
• Amid uncertainty, ad tech mavens turn to startups
• Programmatic snafu speak, translated: a guide to the usual excuses

Editor’s note: This article has been updated to include additional IAS efforts related to CSAM concerns, which were announced Friday evening after this story’s deadline.