WEBVTT

1
00:00:01.300 --> 00:00:08.394
This is Anthropic, Read Aloud. Today: Funding better evaluations of A I’s impact on wellbeing.

2
00:00:09.694 --> 00:00:13.520
Funding better evaluations of A I’s impact on wellbeing.

3
00:00:13.970 --> 00:00:22.300
Anthropic is launching a 5 dollarsmillion grant program to fund independent research into how A I impacts users’ wellbeing.

4
00:00:22.750 --> 00:00:27.386
From Anthropic Newsroom, published August 25, 2026.

5
00:00:28.386 --> 00:00:48.683
We’re launching a 5 dollarsmillion grant program to fund independent research into how A I impacts users’ wellbeing. The program will provide direct funding, access to our models, and technical support to grantees building open-source evaluations that help the A I industry measure how our models affect those who use them.

6
00:00:48.803 --> 00:00:56.122
Grantees will work fully independently, and will publish their work as open-source projects that any developer can make use of.

7
00:00:56.572 --> 00:01:08.369
A I systems have become central to how many people work, learn, and solve problems. They’ve also become conversational partners and can be sources of emotional support during difficult times.

8
00:01:08.489 --> 00:01:23.294
But as an industry, we are still working towards developing clear standards for how models should behave in these conversations, for example, when a user begins to seek companionship from a model, or uses A I to navigate a mental health crisis.

9
00:01:23.745 --> 00:01:37.929
Furthermore, wellbeing is a particularly difficult area to evaluate. For most model behaviors, we can look at a single answer and determine whether it is accurate and appropriate. But assessing wellbeing requires much more context.

10
00:01:38.049 --> 00:01:54.434
For example, a user in distress might not share thoughts of self-harm right away; the need for a more cautious response might only become clear over the course of a long conversation. And a response that might be reasonable in one context might be harmful in another.

11
00:01:54.554 --> 00:02:09.863
For example, Claude might give advice on balanced diets and workout routines to a user who asks about losing weight, but if the user has demonstrated a history of disordered eating, that response could be inappropriate, and potentially actively harmful.

12
00:02:10.313 --> 00:02:29.124
We work to develop safeguards to identify such conversations and help ensure Claude responds appropriately, and we publish research into the types of conversations people have with Claude to better inform how we develop our safeguards, how we evaluate them, and other measures we can take to protect users’ wellbeing.

13
00:02:29.244 --> 00:02:37.905
But these are nuanced considerations, and the stakes are significant. The right approach will need to evolve alongside our models and their uses.

14
00:02:38.355 --> 00:02:53.139
By funding the creation of independent evaluations and benchmarks of user wellbeing, we hope to invite more people to lend their expertise to this emerging and critical field, including clinicians, psychologists, methodologists, and others.

15
00:02:53.939 --> 00:02:57.715
Towards more effective wellbeing evaluations and benchmarks.

16
00:02:58.165 --> 00:03:11.626
As part of this program, we’re sharing guidance from our Safeguards team on what we believe makes a wellbeing evaluation rigorous enough to build on, along with the common challenges that can limit an evaluation’s usefulness.

17
00:03:12.076 --> 00:03:14.345
In brief, we’re seeking evaluations that:

18
00:03:14.795 --> 00:03:20.043
State clearly what they are measuring (that is, what counts as a pass or fail, and why it matters).

19
00:03:20.283 --> 00:03:25.028
Involve clinical and subject-matter experts in the design and validation.

20
00:03:25.268 --> 00:03:31.948
Test both precautions and harms (that is, evaluate the risk of both overcompliance and overrefusal).

21
00:03:32.188 --> 00:03:45.003
Reflect how users actually use A I (often, this means constructing scenarios that represent multi-turn conversations, where risk escalates and context shifts over the course of a long conversation).

22
00:03:45.243 --> 00:03:48.932
Validate their graders against real subject-matter experts.

23
00:03:49.382 --> 00:04:07.702
To learn more about the grant program and apply, see our application form. For more on building strong wellbeing evaluations and benchmarks, read our guidance. Applications are due by September 21; applicants who are selected to submit full proposals will be notified by October 5.

24
00:04:08.702 --> 00:04:26.985
That was Funding better evaluations of A I’s impact on wellbeing, from Anthropic Newsroom, published August 25, 2026. Read the original at anthropic dot com. This unofficial audio edition was generated automatically for the podcast Anthropic, Read Aloud.