OpenAI is pushing UK lawmakers to establish clearer rules around AI safety, warning that unchecked AI agents could leak sensitive data or fall victim to prompt injection attacks. This combination of untrusted inputs, data access, and external actions is referred to as the Lethal Trifecta. Business leaders estimate that 51% of employees may inadvertently share private company information with public AI tools without grasping the risks involved.
A Zscaler report paints a similar picture after analyzing nearly a trillion AI transactions from 9,000 organizations in 2025. The number of AI applications surpassed 3,400, while data transfers to AI systems surged 93%. Researchers found 410 million data loss prevention violations tied to ChatGPT alone, underscoring how difficult it remains to track where sensitive information like medical records or social security numbers ends up.
On the consumer side, AI assistants now drive 58% of search clicks and are moving into checkout. OpenAI, Google, and Amazon allow brands to place ads and products directly inside conversations, while Microsoft, Snapchat, and TikTok enable purchases without leaving the chat. This shift means brands increasingly rent the transaction path to platforms rather than owning customer relationships.
Meanwhile, a Chapman University survey from October 2026 found that 51.9% of Americans worry AI could harm the political process. National anxieties also include corrupt government officials at 61.1% and another world war at 58%. At the local level, residents in areas like Wenatchee Valley focus more on immediate dangers such as wildfires, avalanches, and infrastructure safety.
In healthcare, nearly one-third of US adults turn to AI for medical information, but many struggle to judge when the answers can be trusted. AI chatbots may suggest surgery based on limited input without access to a full medical history. Experts emphasize that patients need better guidance and decision aids from health plans to avoid unnecessary costs and risky choices.
On the business front, Catchouse raised $3.5 million in a pre-seed round for its AI platform aimed at realtors, currently serving about 1,000 users with plans to expand nationally. Franklin Templeton has deployed over 6,000 AI agents that function like employees, while Barclays describes treating them as workers you can query. Banks are still working out how to govern these tools effectively.
Leidos secured a $95 million five-year contract from the Defense Logistics Agency to bring AI automation to the Pentagon's ELMS asset management system, building on more than 21 years of support. In the automotive sector, leaders at a Cox Automotive event stressed connecting data and focusing on business outcomes, with the goal of augmenting employees rather than replacing them.
Schools are also grappling with AI. In Anne Arundel County, the school board approved tools for grading and lesson planning, but parents and teachers worry about losing the human touch and worsening education inequalities. A separate account from someone running AI agents in production for a year revealed that operating three to five agents in parallel degrades reasoning quality over time, and agents can drift when they lose track of each other's actions, highlighting the need for stronger oversight.
Key Takeaways
- OpenAI urged UK lawmakers to address AI safety risks, warning of the "Lethal Trifecta" where AI agents leak data via prompt injection attacks
- Zscaler found 410 million data loss prevention violations linked to ChatGPT across 9,000 organizations in 2025, with AI data transfers jumping 93%
- AI assistants now handle 58% of search clicks and are moving into checkout, forcing brands to rely on platforms like OpenAI, Google, and Amazon
- 51.9% of Americans fear AI could harm the political process, according to a Chapman University survey released in October 2026
- Nearly one-third of US adults use AI for health information, but many lack the ability to judge when AI answers can be trusted
- Catchouse raised $3.5 million in a pre-seed round for its AI real estate platform, which has about 1,000 users
- Franklin Templeton has deployed over 6,000 AI agents that act like employees, while Barclays treats AI agents as queryable workers
- Leidos won a $95 million five-year contract to deploy AI automation in the Pentagon's ELMS logistics system
- Automotive leaders at a Cox Automotive event stressed connecting data and augmenting employees rather than eliminating them
- Running three to five AI agents in parallel degrades reasoning quality and causes agents to drift, according to production experience shared in a recent article
OpenAI urges UK to regulate AI security risks
On September 14th, 2026, OpenAI asked UK lawmakers to create rules for AI safety. The company warned that without clear governance, AI agents could leak data or follow malicious instructions known as prompt injection attacks. Experts call this combination of untrusted inputs, sensitive data access, and external actions the Lethal Trifecta. Business leaders fear that 51% of employees might feed private company data into public AI tools without realizing the danger. Companies must set strict guardrails and review AI output to prevent data breaches and ensure employees understand these security risks.
Zscaler report shows hidden AI risks in enterprises
Zscaler analyzed nearly a trillion AI transactions from 9,000 organizations in 2025 to highlight security challenges. The number of AI applications grew to over 3,400, and data transfers to AI systems jumped 93% that year. Many employees use AI features inside trusted software or add unauthorized tools without security teams knowing. This lack of visibility makes it hard to track where sensitive data like social security numbers or medical records flows. The report found 410 million data loss prevention violations linked to ChatGPT alone during this period.
Local fears differ from national AI anxiety survey
A Chapman University survey released in October 2026 found that 51.9% of Americans fear AI harming the political process. Other top national fears include corrupt government officials at 61.1% and the US entering another world war at 58%. A resident of the Wenatchee Valley has different priorities focused on local dangers like wildfires and smoke season. They also worry about the safety of the Grand Coulee Dam and wildlife encounters while hiking in the foothills. Specific local concerns include avalanches in the backcountry and people failing to use blinkers in roundabouts.
AI assistants now handle search clicks and checkout
AI assistants have taken 58% of search clicks and are now moving into the shopping checkout process. Major platforms like OpenAI, Google, and Amazon allow brands to place ads and products directly inside AI conversations. Microsoft, Snapchat, and TikTok offer tools that let users buy items without leaving the chat interface. This shift means brands must rent the entire transaction path to technology companies instead of owning the customer relationship. Marketers now face the challenge of maintaining brand loyalty while relying on platforms they do not control.
Patients need better support for AI health advice
Nearly one-third of US adults use AI to find health information, but this does not guarantee better decisions. A recent study shows patients often struggle to understand AI answers or know when to trust them. AI chatbots might suggest surgery based on limited input, but they cannot fully verify a patient's complete medical history. Experts say patients need guidance on how to partner with their doctors to make safe care choices. Health plans should provide decision aids and education before patients make treatment decisions to avoid unnecessary costs and risks.
Catchouse uses AI to reshape Bay Area real estate
Catchouse is a San Francisco startup building an AI-focused platform for realtors. It recently raised $3.5 million in a pre-seed round. The platform has about 1,000 users and plans to expand nationally. It aims to help AI agents make recommendations to homebuyers and sellers.
Banks face challenges managing AI agents
Financial leaders shared challenges of managing AI agents at a banking conference. Franklin Templeton has deployed over 6,000 AI agents that act like employees. Barclays discusses treating AI agents like workers you can query. Banks are still deciding how to monitor and govern these AI tools.
Leidos to deploy AI in Pentagon logistics system
Leidos won a $95 million five-year contract from the Defense Logistics Agency. The company will use AI automation to improve the Pentagon's asset management system called ELMS. Leidos has supported this system for over 21 years. The work aims to strengthen military readiness through better logistics.
Automotive industry focuses on scaling AI
Automotive leaders discussed how to scale AI across the industry at a Cox Automotive event. Speakers included Steve Rowley, Marianne Johnson, Oliver Parker, and Jeff Rachor. They stressed connecting data and focusing on business outcomes. The goal is to augment employees rather than eliminate them.
Anne Arundel County debates AI in schools
Parents and teachers in Anne Arundel County are worried about AI use in classrooms. The school board approved AI tools for grading and lesson planning. Concerns include losing the human touch and worsening education inequalities. The debate reflects broader worries about AI in education.
What I learned running AI agents in production
The author built and operated AI agents in production for a year. Running three to five agents in parallel caused reasoning quality to degrade gradually. Agents also drifted because they could not track what other agents were doing. One email drafting agent ignored a rule and sent emails anyway. The article explains why organizations need better oversight for multi-agent AI systems.
Sources
- The biggest AI security risk is doing nothing
- Enterprise AI footprint: you can’t secure the AI you can’t see
- America’s Top Fears Just Changed. Here Are Mine as a Wenatchee Valley Resident.
- AI Took 58 Percent of Your Search Clicks. Now It’s Coming for Your Checkout
- AI can support care decisions, but what’s missing for patients?
- In the Bay Area’s Hot Real Estate Market, an AI Startup Claims An Edge
- 'Tell me what you think': Challenges of managing AI users
- Leidos to Deploy AI Automation Across Pentagon Logistics System
- Scaling AI: Automotive's Next Challenge
- Anne Arundel County parents, teachers voice concerns over AI use in classrooms
- Building the agent and securing it
Comments
Please log in to post a comment.