Anthropic Is Letting Social Workers From Hundreds Of Government Agencies Use Its AI For Paperwork

Anthropic is partnering with Binti, a startup that helps social workers manage their cases.

SOPA Images/LightRocket via Getty Images

Paperwork is one of the biggest stumbling blocks to meaningful social work, the kind that places at-risk kids in loving families. The process of gathering and filling out forms for court documents, family assessments, referrals for housing programs and more delays the actual work of supporting and improving the wellbeing of those in need.

Anthropic thinks AI can help. The company has partnered with Binti, a startup providing a digital platform for more than 12,000 social workers across some 550 government social service agencies, to build out new tools to help them manage their cases more easily.

One feature will transcribe recordings of social workers’ meetings with families (with their consent), and use that information to fill in paperwork like administrative reports. The social worker can also upload handwritten case notes that the AI can use to fill in forms. They must then approve the AI-generated draft or make changes, to ensure the software didn’t insert any errors. Another feature will let social workers upload everything related to a case, including forms, notes, case data and agency policies. Then the worker can “chat” with their case to ask it questions and get quick answers without having to sort through hundreds of pages of documents.

“Social workers got into this field to help children and families,” Binti CEO and cofounder Felicia Curcuru told Forbes. “And they hate the paperwork, universally.”

The partnership is Anthropic’s first foray into government social work, an unglamorous sector that normally doesn’t receive much attention, let alone high-tech tools. It’s part of Anthropic’s AI for Social Good program, an initiative launched by the company in May that aims to provide AI tools and support to companies and nonprofits that are working toward societal impact, but might not be as buzzy or commercially successful as other startups. “When we founded Anthropic, we believed AI could transform lives in meaningful ways,” CEO Dario Amodei said in a statement. “Seeing Binti help connect children with loving families faster by cutting through months of paperwork is exactly the real-world impact we hoped to enable.”

AI has already started impacting industries ranging from healthcare and law to customer service and computer science. But Anthropic’s interest in government social work illustrates its ambition for an even wider scope. That could impact the labor force, which Amodei has not been shy about addressing. Earlier this year, he said AI could wipe out half of all entry-level white-collar jobs and spike unemployment up to 20% within the next five years.

Asked about the possibility of Claude eventually taking the job of social workers, Elizabeth Kelly, head of Anthropic’s AI for Social Good program, emphasized the value of “human connection.” “It’s important to be clear that AI is about enabling people,” she told Forbes.

“Social workers got into this field to help children and families. And they hate the paperwork, universally.”

Felicia Curcuru

Kelly, a former Biden administration official who previously ran the U.S. AI Safety Institute, declined to disclose the exact terms of the partnership with Binti. But she said members of the program can typically receive token credits, marketing resources, help with go-to-market strategies and access to Anthropic’s technical team, which can assist with things like prompt engineering and fine tuning.

The AI for Social Good program’s other high-profile partnerships include providing free processing power and technical support to NextLadder Ventures, a joint entity from the charitable foundations of billionaires Bill Gates, Charles Koch, Steve Ballmer, Intuit founder Scott Cook and hedge fund investor John Overdeck. The project, which launched in July, will support organizations focused on using AI and other emerging tech to improve the financial trajectory of low-income Americans.

As for using AI for social work, there are pitfalls to watch for. AI in general has been known to make things up, or “hallucinate,” when it doesn’t have the answers. Its training data can also have an outsize impact on its responses, leading to the potential for bias. Curcuru said Binti has tried to mitigate those risks by turning down the “creativity” of Anthropic’s Claude in this instance, so it will only pull from the data it’s given, and not guess or make inferences.

Some experts are skeptical. “These tools are optimized for plausibility. Inaccuracy is a known problem that is not going away,” said Clara Berridge, a professor at the University of Washington School of Social Work who researches AI’s impact on the job. “So how much time is saved by using the tool if the social worker is actually reading through and checking everything coming from the tool to ensure the information is correct?”

While Curcuru said she’s received positive feedback from social workers on the AI features, she noted one of the biggest roadblocks so far has been the hesitance of some social workers to use the tool at all. Some workers will be reticent to ask families for their permission to record meetings, thinking it will make them uncomfortable. But Curcuru said that when workers do ask, the majority of families are “totally okay with it.” “It can really streamline their administrative work so they focus on the sort of work they got into the field for.”

Binti launched in 2017 with the goal of modernizing the management of social work. Curcuru has long had a personal connection to child welfare — she has volunteered with foster youth for a decade, and had a front row seat to the challenge and stress of adoption when her sister adopted two children. After shadowing child welfare social workers in San Francisco County, Curcuru noticed the rudimentary tools they used to keep track of families, including massive Excel spreadsheets and Post-It notes, and decided to do something about it. Now, Binti boasts that more than 12,000 social workers use its tech, saving up to 40% of their time on administrative tasks.

The company works with agents across 36 states that serve almost half of all child welfare in the U.S., including the Los Angeles County Department of Children and Family Services and Utah Foster Care. Binti has raised some $60 million in venture capital — modest compared to the eye-popping funding rounds of AI startups — from backers including Founders Fund (its largest investor, led by partner Trae Stephens), First Round Capital and Kapor Capital. And in March, it got the attention of Anthropic CEO Amodei.

Curcuru met the CEO in March at an Allen and Company conference in Arizona. As the two chatted over lunch, Amodei told her he was looking for ways for Anthropic to work closely with governments. Curcuru saw an opening. Amodei also wasn’t interested in training Claude, the company’s popular large language model, or LLM, on sensitive patient information, Curcuru recalls him telling her. “It made me want to lean into working with Anthropic over other LLMs,” she says now.