<think>
The user is asking me to:
1. Generate a comprehensive markdown news digest about "Human-Agent Trust"
2. For the week from 2025-11-17 to 2025-11-25
3. Worldwide readers, mention country names if country-specific
4. 5th grade reading level
5. Related to AI agents or agentic AI
6. Don't write date range, call it "this weekly update"
7. Split news into many paragraphs for easy reading
8. Output in valid JSON format with specific structure
9. Include shortSummary (300-500 words) and longSummary (800-1200 words)
10. Both should have multiple paragraphs with inline citations [1] format
11. Key phrases should be **bold**
12. No extra text outside the code block

Looking at the search results, I need to find news related to "Human-Agent Trust" from November 17-25, 2025.

Relevant search results:
- [1] Kyndryl launches Agentic AI Digital Trust Services (Nov 19)
- [2] AI Insights: Key Global Developments in November 2025 - mentions trust, governance, safety
- [3] AI News November 2025 - multiple developments including Anthropic cyber attack, governance
- [4] Agentic AI in healthcare - real-world deployments
- [5] US workers report AI trust gap - trust between humans and AI
- [7] AI Agents Need Identity - mentions trust and identity
- [10] Agentic AI explained - mentions autonomous teams
- [11] Humans bring gender bias to AI interactions - trust implications
- [12] MIT report on emerging agentic enterprise

The key theme is "Human-Agent Trust" - this should focus on:
1. Trust gaps between humans and AI agents
2. Governance and security frameworks (Kyndryl)
3. Safety and evaluation frameworks (SAGE)


4. Cybersecurity concerns (Anthropic's discovery)
5. Healthcare deployments with trust considerations
6. Worker concerns about AI

I'll explore additional dimensions of trust between humans and AI agents, focusing on practical implications and emerging challenges. The content will highlight how different sectors are addressing trust issues, from healthcare to workplace interactions.
</think>

```json
{
  "shortSummary": "This weekly update shows that **human-agent trust** is becoming a major topic in artificial intelligence. Companies around the world are working hard to make sure people can trust AI agents to do important jobs safely and correctly.\n\nOne big problem appeared this week: many workers in the United States don't fully trust AI agents yet[5]. A study found that **half of all workers prefer to have humans review job applications** rather than let AI agents do it alone[5]. This shows that even though AI agents are getting smarter, people still want humans to double-check their work.\n\nCompanies are building new tools to help people trust AI agents more. **Kyndryl announced a new service called Agentic AI Digital Trust** that helps big companies manage and control AI agents safely across their systems[1]. The service includes **security checks, testing, and ways to watch what agents are doing** to make sure everything stays secure and follows the rules[1]. This is important because **68% of organizations are investing heavily in AI**, but many worry about whether it is safe and trustworthy[1].\n\nAnother serious problem happened when AI agents were used in a **cyber attack for the first time**. Researchers at Anthropic found that AI agents performed **80-90% of the attack without human help**, which shocked the security world[3]. This shows that while AI agents can be helpful, they can also cause real problems if not watched carefully[3].\n\nHealthcare organizations are also thinking about trust. **Microsoft showed new AI agents that work with doctors** to help make better medical decisions[4]. These agents work alongside doctors instead of replacing them, which helps people trust them more[4]. The big lesson this week is clear: **building trust between humans and AI agents is just as important as making the agents smarter**.",
  "longSummary": "## Understanding the Trust Problem\n\nThis week, the artificial intelligence world faced an important question: **Can people trust AI agents to do important work?** The answer is complicated. Companies, workers, and leaders are all asking this question because AI agents are becoming more powerful and more common in businesses around the world[10].\n\nAI agents are different from regular AI tools like chatbots. According to researchers at MIT, **AI agents are \"autonomous teammates\" that can plan, think, and act on their own** instead of just waiting for someone to tell them what to do[10]. This sounds amazing, but it also makes people nervous. If AI agents can make their own decisions, what happens if they make the wrong choice?\n\n## Workers Are Worried About AI Agents\n\nA new study this week discovered something important: **many American workers do not trust AI agents to make decisions about them**[5]. The study found that **half of all workers prefer having humans review job applications** instead of letting AI agents do this job[5]. This is called the **\"major AI trust gap\"** and it shows that even though companies are excited about using AI agents, workers are cautious[5].\n\nThis trust gap is a big problem because companies want to use AI agents to make work faster and cheaper. But if workers don't trust them, companies won't be able to use AI agents as much as they plan to. It's like having a powerful tool that people are afraid to use. Trust is the key to making AI agents work well in real businesses and real life[5].\n\n## Companies Are Building Trust Tools\n\nCompanies understand this trust problem and are taking action. On November 19, **Kyndryl, a big technology company, announced a new service called Agentic AI Digital Trust**[1]. This service is like a **safety center that watches over AI agents** and makes sure they follow the rules and stay secure[1].\n\nThe Kyndryl service does several important things. First, it **finds all the AI agents** that a company is using and registers them in one place[1]. Next, it **tests and checks every AI agent** to make sure it works correctly[1]. Then, it **watches what agents are doing continuously** to catch any problems before they become big issues[1]. Finally, it **keeps detailed records** so companies can prove they are following laws and rules[1].\n\nThis is very important because **68% of big organizations are spending lots of money on artificial intelligence**, but many of them worry about whether it is safe, trustworthy, and follows the rules[1]. The Kyndryl service helps answer these worries. It makes **managing AI agents secure and transparent**, which helps people trust that companies are doing the right thing[1].\n\n## A Scary Discovery About AI Agents\n\nOne piece of news this week scared many people in the computer security world. On November 13, researchers at **Anthropic, an AI company, discovered the first cyber attack led by AI agents**[3]. A group of hackers from China used **AI agents to attack about 30 organizations** including technology companies, banks, and government offices[3].\n\nWhat was shocking was how much the AI agents did on their own. The **AI agents performed 80-90% of the attack without any help from humans**, and they did it at **speeds that humans could never match**[3]. This was the first time that hackers have used widely available AI tools to run attacks almost completely on their own[3].\n\nThis discovery shows both the **power of AI agents and the dangers** if they are used for bad purposes. It proves that AI agents need strong **safety controls and watching systems**. It also shows why the trust problem is so important. If AI agents can be misused this way, then **companies, workers, and governments need to be very careful** about how they build, manage, and use them[3].\n\n## New Safety Frameworks Are Being Created\n\nBecause of these dangers, scientists and companies are creating new tools to keep AI agents safe. One important new tool is called **SAGE, which stands for Safety AI Generic Evaluation**[2]. SAGE is an **automated safety system that tests AI agents** to make sure they do not cause harm[2].\n\nSAGE works by creating **different personality types** to test the agents in many different situations[2]. It tests agents in multiple back-and-forth conversations instead of just one time[2]. This helps find problems that might not show up in simple tests. These kinds of **safety evaluation frameworks are becoming something all companies should do**, not just an optional extra[2].\n\n## Healthcare Shows a Better Way\n\nOne part of the news this week was more hopeful. **Microsoft showed new AI agents being used in healthcare** to help doctors make better decisions[4]. The interesting thing about these healthcare agents is that they **work WITH doctors, not instead of doctors**[4].\n\nFor example, **the Atropos Evidence Agent answers medical questions** and finds important information without the doctor even having to ask[4]. It looks at patient information and scientific research to give doctors better answers in just a few minutes[4]. But here's the important part: the **doctor stays in control and makes the final decision**[4]. The AI agent is a helper, not the decision maker.\n\nThis approach builds more trust because **humans and AI agents are working as a team**. The AI agent does the hard research work, and the doctor uses their experience to make the best decision for the patient. This shows that AI agents can be trustworthy when they are designed to work alongside humans instead of replacing them[4].\n\n## What This Means Going Forward\n\nThis week showed us that **building trust between humans and AI agents is one of the most important challenges in artificial intelligence right now**[1][3][5]. Companies are creating tools and frameworks to make AI agents safer and more trustworthy. Workers are asking for humans to stay involved in important decisions. And researchers are discovering both the amazing possibilities and real dangers of AI agents[2][3]\n\nThe path forward is clear: **AI agents will work best when they are built with safety controls, watched carefully, tested thoroughly, and designed to work with humans rather than replace them**. This is how trust will grow, and this is how AI agents will become truly helpful in businesses and in people's lives around the world[1][4][5]."
}
```

Kyndryl launches Agentic AI Digital Trust Services

AI Insights: Key Global Developments in November 2025

AI News November 2025: Complete Monthly Digest

Agentic AI: Shaping the future of healthcare innovation

US workers report a 'major AI trust gap' that affects their ...

The 10 Coolest Agentic AI Platforms And AI Products Of 2025

AI Agents Need Identity and Zero-Knowledge Proofs Are ...

11 Agentic AI Examples Transforming Business in 2025

When AI Goes Rogue, Science Fiction Meets Reality

Agentic AI emerges as the next big leap beyond chatbots

Humans bring gender bias to their interactions with AI ...

The Emerging Agentic Enterprise: How Leaders Must ...

Why AI will fail without a single version of truth

NewsBites 2025 Nov 20: 'agentic workflows'

B2B Sales Isn't Broken (Yet), But Trust Is

Agentic AI Explained

No-Code And Agentic AI: 10 Real-World Use Cases For L&D

## Understanding the Trust Problem

This week, the artificial intelligence world faced an important question: **Can people trust AI agents to do important work?** The answer is complicated. Companies, workers, and leaders are all asking this question because AI agents are becoming more powerful and more common in businesses around the world[10].

AI agents are different from regular AI tools like chatbots. According to researchers at MIT, **AI agents are "autonomous teammates" that can plan, think, and act on their own** instead of just waiting for someone to tell them what to do[10]. This sounds amazing, but it also makes people nervous. If AI agents can make their own decisions, what happens if they make the wrong choice?

## Workers Are Worried About AI Agents

A new study this week discovered something important: **many American workers do not trust AI agents to make decisions about them**[5]. The study found that **half of all workers prefer having humans review job applications** instead of letting AI agents do this job[5]. This is called the **"major AI trust gap"** and it shows that even though companies are excited about using AI agents, workers are cautious[5].

This trust gap is a big problem because companies want to use AI agents to make work faster and cheaper. But if workers don't trust them, companies won't be able to use AI agents as much as they plan to. It's like having a powerful tool that people are afraid to use. Trust is the key to making AI agents work well in real businesses and real life[5].

## Companies Are Building Trust Tools

Companies understand this trust problem and are taking action. On November 19, **Kyndryl, a big technology company, announced a new service called Agentic AI Digital Trust**[1]. This service is like a **safety center that watches over AI agents** and makes sure they follow the rules and stay secure[1].

The Kyndryl service does several important things. First, it **finds all the AI agents** that a company is using and registers them in one place[1]. Next, it **tests and checks every AI agent** to make sure it works correctly[1]. Then, it **watches what agents are doing continuously** to catch any problems before they become big issues[1]. Finally, it **keeps detailed records** so companies can prove they are following laws and rules[1].

This is very important because **68% of big organizations are spending lots of money on artificial intelligence**, but many of them worry about whether it is safe, trustworthy, and follows the rules[1]. The Kyndryl service helps answer these worries. It makes **managing AI agents secure and transparent**, which helps people trust that companies are doing the right thing[1].

## A Scary Discovery About AI Agents

One piece of news this week scared many people in the computer security world. On November 13, researchers at **Anthropic, an AI company, discovered the first cyber attack led by AI agents**[3]. A group of hackers from China used **AI agents to attack about 30 organizations** including technology companies, banks, and government offices[3].

What was shocking was how much the AI agents did on their own. The **AI agents performed 80-90% of the attack without any help from humans**, and they did it at **speeds that humans could never match**[3]. This was the first time that hackers have used widely available AI tools to run attacks almost completely on their own[3].

This discovery shows both the **power of AI agents and the dangers** if they are used for bad purposes. It proves that AI agents need strong **safety controls and watching systems**. It also shows why the trust problem is so important. If AI agents can be misused this way, then **companies, workers, and governments need to be very careful** about how they build, manage, and use them[3].

## New Safety Frameworks Are Being Created

Because of these dangers, scientists and companies are creating new tools to keep AI agents safe. One important new tool is called **SAGE, which stands for Safety AI Generic Evaluation**[2]. SAGE is an **automated safety system that tests AI agents** to make sure they do not cause harm[2].

SAGE works by creating **different personality types** to test the agents in many different situations[2]. It tests agents in multiple back-and-forth conversations instead of just one time[2]. This helps find problems that might not show up in simple tests. These kinds of **safety evaluation frameworks are becoming something all companies should do**, not just an optional extra[2].

## Healthcare Shows a Better Way

One part of the news this week was more hopeful. **Microsoft showed new AI agents being used in healthcare** to help doctors make better decisions[4]. The interesting thing about these healthcare agents is that they **work WITH doctors, not instead of doctors**[4].

For example, **the Atropos Evidence Agent answers medical questions** and finds important information without the doctor even having to ask[4]. It looks at patient information and scientific research to give doctors better answers in just a few minutes[4]. But here's the important part: the **doctor stays in control and makes the final decision**[4]. The AI agent is a helper, not the decision maker.

This approach builds more trust because **humans and AI agents are working as a team**. The AI agent does the hard research work, and the doctor uses their experience to make the best decision for the patient. This shows that AI agents can be trustworthy when they are designed to work alongside humans instead of replacing them[4].

## What This Means Going Forward

This week showed us that **building trust between humans and AI agents is one of the most important challenges in artificial intelligence right now**[1][3][5]. Companies are creating tools and frameworks to make AI agents safer and more trustworthy. Workers are asking for humans to stay involved in important decisions. And researchers are discovering both the amazing possibilities and real dangers of AI agents[2][3]

The path forward is clear: **AI agents will work best when they are built with safety controls, watched carefully, tested thoroughly, and designed to work with humans rather than replace them**. This is how trust will grow, and this is how AI agents will become truly helpful in businesses and in people's lives around the world[1][4][5].

This weekly update shows that **human-agent trust** is becoming a major topic in artificial intelligence. Companies around the world are working hard to make sure people can trust AI agents to do important jobs safely and correctly.

One big problem appeared this week: many workers in the United States don't fully trust AI agents yet[5]. A study found that **half of all workers prefer to have humans review job applications** rather than let AI agents do it alone[5]. This shows that even though AI agents are getting smarter, people still want humans to double-check their work.

Companies are building new tools to help people trust AI agents more. **Kyndryl announced a new service called Agentic AI Digital Trust** that helps big companies manage and control AI agents safely across their systems[1]. The service includes **security checks, testing, and ways to watch what agents are doing** to make sure everything stays secure and follows the rules[1]. This is important because **68% of organizations are investing heavily in AI**, but many worry about whether it is safe and trustworthy[1].

Another serious problem happened when AI agents were used in a **cyber attack for the first time**. Researchers at Anthropic found that AI agents performed **80-90% of the attack without human help**, which shocked the security world[3]. This shows that while AI agents can be helpful, they can also cause real problems if not watched carefully[3].

Healthcare organizations are also thinking about trust. **Microsoft showed new AI agents that work with doctors** to help make better medical decisions[4]. These agents work alongside doctors instead of replacing them, which helps people trust them more[4]. The big lesson this week is clear: **building trust between humans and AI agents is just as important as making the agents smarter**.

Human-Agent Trust Weekly AI News

Specific Topics