Meta announced Wednesday that it took action against 33.2 million pieces of child sexual exploitation content on Facebook and Instagram in the first half of 2026. At the same time, the company is rolling out new AI tools to find ads and accounts that may be used to secretly direct people to illegal child abuse material.

More than 97% of the content Meta acted on was found by its systems before users reported it, the company said. In India, Meta acted on 5.3 million pieces of child sexual exploitation content during the same period, with more than 98% detected before users reported it.

Meta said its latest AI tools are aimed at people using ads that look harmless to send users to illegal content outside Meta’s platforms.

The company has introduced a new large language model (LLM) system to detect what it calls “signposting.” This refers to ads that may look normal but are suspected of directing users to illegal content or other harmful activity elsewhere online. Meta says this is a tactic it has recently seen bad actors use as they continue to change their methods to avoid detection. While the ads themselves may not contain illegal material, they can still act as a gateway to websites hosting child sexual abuse material.

Because of this, Meta said it’s now looking at where an ad sends users, and not just what the ad contains. The company can use this information to block websites or other destinations that break its rules and take action against the accounts behind them.

Meta is also using additional AI-driven scans to find child exploitation content that earlier systems may have missed. The company said it will continue adding new signals as it learns more about how these networks operate.

Another new tool is a “red-teaming AI agent” that tests Meta’s own safety measures. It looks for weaknesses that bad actors could use to get around the company’s protections. Meta said this could help it find new methods of abuse before they become more common.

Additionally, the company is improving its systems for finding people who return to its platforms with new accounts after their previous accounts have been removed.

The new measures come as Meta continues to face pressure over the safety of children on its platforms. The company has faced lawsuits and criticism from lawmakers over the risks its social media services may pose to young users. In August, Meta reached an agreement to pay up to $18 billion to settle a child safety lawsuit involving 29 U.S. states, addressing concerns regarding the protection of children online.

Meta has introduced several child-safety features this year, including parental controls for Meta AI, pre-teen accounts on WhatsApp, and alerts for parents when children search Instagram for self-harm content.

In September, WhatsApp added more controls for parents. They can limit how teenagers use Channels, control who can see their status updates, and decide who can add their children to groups. Parents can also choose to receive notifications about certain group activity.

More Information

Humanoid robots must learn to balance, move, and interact with objects across an enormous range of situations, but the data available for this training has clear limits. Internet video shows diverse behavior but cannot capture precise physical states. Laboratory motion capture systems record accurate movement but usually cover only a narrow set of actions. This white paper examines HiPHI, a 617.5-hour whole-body human motion dataset captured with optical motion capture at sub-millimeter accuracy. It includes 245.7 hours of human-object interaction with synchronized object trajectories and meshes, and organizes coverage using FrameNet, a linguistic framework for human action. The paper also introduces a benchmark suite for measuring motion diversity and interaction grounding, and reports results from policies trained on the dataset and deployed on a physical Unitree G1 humanoid robot.

When Justin Kosmides co-founded Bloom in 2023, his goal was to create a savior for the mobility industry. He had seen many e-bike and e-scooter companies fail because of an inability — or an unwillingness — to outsource some of the hardest problems, like logistics, manufacturing, or building a supply chain. He wanted to convince survivors and new entrants to let Bloom handle that work for them.

Then Donald Trump got re-elected and started shotgun-blasting tariffs on dozens of countries, partially to jump-start U.S. manufacturing. This accelerated a hardware scene that was already heating up in this country and, suddenly, it was more than just mobility companies looking to beef up their domestic supply chain. Robotics startups, drone-makers, and many more started to pop up.

That was an opportunity for Bloom, which is based in Detroit, but to tackle it, the startup had to partially reinvent itself. Instead of performing some of those harder, behind-the-scenes tasks, which was a big part of the original plan, Bloom shifted to a pure marketplace model, in which it connects buyers and sellers. It’s now focused on making, essentially, supply chain AI agents that Bloom’s customers use to find specific types of suppliers, parts, or manufacturing and engineering services.

The reinvention slowed Bloom’s fundraising plans. But the company has now made over 2,000 matches for more than 140 companies, and it wants grow faster. In an interview earlier this year, Kosmides, who is Bloom’s CEO, said his startup is like AI-driven version of Alibaba did in China, where it created a marketplace for contract manufacturers.

Investors are finally buying in. On Wednesday, Bloom announced a $3.6 million seed round led by SNAK Venture Partners, a marketplace-focused investment firm. Also joining were Flyover Capital (an early stage investor that focuses on flyover states), and deep tech firm Mana Ventures. Local backers included Detroit Venture Partners, Invest Detroit Ventures, and the Michigan Outdoor Innovation Fund.

Kosmides, sounding relieved, said in the interview that he’s “excited to be done with fundraising and get to back to building.”

No, then yes

SNAK first met with Bloom in April of last year, in the early stages of this reinvention. So early, in fact, that the firm passed on the chance to invest in the startup’s pre-seed round.

“We liked the founder and the thesis and passed anyway: we wanted to see more traction,” the firm wrote in a blog post. “We said so plainly and kept tracking the company.”

Bloom kept in touch with SNAK as it worked through becoming a more software-focused startup. By May of this year, Bloom had made as much revenue in five months as it did in all of 2025. SNAK also said the number of memberships on its platform had grown fivefold “with low churn,” meaning few customers were canceling.

“This is meant to reiterate that for us, a pass on pre-seed is not always a pass forever. We are fortunate to have a super focused thesis that allows us to track a small set of interesting early stage companies and build a relationship over time,” the firm wrote.

Kosmides said the fundraising scene has been a bit ruthless for startups like Bloom. Proving your value to investors in a world where the leading AI models keep improving can be an uphill battle.

“To find investors to actually write a term sheet, not follow one, is getting harder and harder,” he said. “That’s just this new era that we’re in, with everyone trying to figure out what’s real and what’s not, what to invest in.”

Bloom did court other term sheets, Kosmides said. But SNAK, which was founded by longtime retail executive Sonia Nagar, is “who you want in your camp” to “build a hopefully category-defining marketplace for mobility, and drones, and all this hardware.”

Matchmaking and discovery

Bloom’s platform handles both sides of a supplier relationship. Some customers use the platform to post contract opportunities, while others use it to bid on those opportunities. Bloom handles quoting, booking, and payment all on its platform, and it now helps match companies with suppliers that can do contract manufacturing, assembly, design and engineering, freight, warehousing, repairs, and even hazardous-materials shipping.

“Our friends at the parts marketplaces like Fictiv, Xometry, MacroFab and others are really good yesteryear marketplaces for finding one particular part,” Kosmides said. “But when you’re XYZ Drone Company, or an electric motorcycle company, and you need to find a provider that meets all these requirements have them all bid on this opportunity, that’s a lot more complicated than finding a CNC part.”

Making this system work well is a big part Bloom’s reinvention, and it involved ingesting a lot of data about each provider on the platform. A lot of that was public data — how a company describes itself on its website, for instance. But around 30-40% of that data is coming directly from the companies, Kosmides said. Add in the data being generated as the startup onboards more customers and makes more matches, and you have Bloom’s edge, he said.

“You can build the craziest model for scraping data but you would never get to that accuracy,” he said. “We started with manually booking services, and then built in layers and pieces and understanding as we built up the provider network, as we built out more and more transactions, and so it’s only getting better and better.”

Kosmides believes this is all a better way to strike up supplier relationships, and that it also makes it easier for startups to get off the ground. But he also thinks Bloom’s platform will help create opportunities for small manufacturers who may not have big marketing budgets or sales teams.

He gave an example of one Michigan-based contract manufacturer that, until it started working with Bloom, was taking on odd jobs like refurbishing Nest thermostats and Bird scooters, or working with Chick-fil-A displays. That same company is now bidding on drone assembly contracts.

“The ability to discover, and the ability to do matchmaking, that is what I think is truly missing” from U.S. manufacturing, he said. “That just doesn’t exist electronically.”