Salesforce Fixes Vulnerability, Google Joins UK AI Health Commission

The artificial intelligence landscape is rapidly evolving, marked by significant advancements, emerging security concerns, and global regulatory efforts. Salesforce is addressing a critical vulnerability in its Agentforce AI platform, dubbed 'ForcedLeak,' which allowed AI agents to be tricked into exposing sensitive customer data through prompt injection attacks. The company has since implemented fixes, including securing an expired domain and blocking output to untrusted URLs. Meanwhile, the UK is taking a proactive approach to AI in healthcare, establishing a national commission to guide the safe adoption of AI in the NHS. This commission, involving experts and consulting with companies like Google, aims to update regulations and foster innovation, with a framework expected in 2026. On a broader scale, the United Nations is grappling with the global implications of AI, launching a dialogue on governance and warning of potential risks, particularly concerning autonomous weapons. One of ChatGPT's co-creators, Illia Polosukhin, has voiced concerns that current AI systems are 'broken,' advocating for a blockchain-based solution to ensure user ownership and privacy. The economic side of AI is also under scrutiny, with discussions around a potential stock market bubble driven by high valuations and concentrated market leadership. Rising Treasury yields pose a threat to the AI investment boom, potentially increasing the cost of AI infrastructure projects. In education and workforce development, the University of Louisiana Monroe received a grant for an AI cybersecurity education project, and Interview Kickstart launched a program for ML engineers focused on custom AI chips.

Key Takeaways

• Salesforce has fixed a 'ForcedLeak' vulnerability in its Agentforce AI platform that could expose sensitive customer data.
• The UK has formed a national AI health commission to accelerate safe AI adoption in the NHS, with input from companies like Google.
• A UN initiative is underway to establish global governance for AI, addressing risks and promoting inclusive development.
• Illia Polosukhin, a co-creator of the foundational AI paper, believes current AI is 'broken' and proposes a blockchain solution for user-owned AI.
• Concerns exist about a potential AI stock market bubble due to high valuations and concentrated market leadership.
• Rising Treasury yields may hinder the AI investment boom by increasing the cost of infrastructure projects.
• The University of Louisiana Monroe received a $129,532 grant for an AI cybersecurity education project.
• Interview Kickstart launched a Machine Learning Engineer program focused on AI for custom silicon chips.
• New partnerships are emerging for AI-driven solutions, such as virtual product placement in content.
• Generation Z is influencing workplace culture, while significant investments are being made in AI data centers.

Salesforce AI Agents Exposed Sensitive Data Via 'ForcedLeak' Flaw

Researchers discovered a critical vulnerability in Salesforce's Agentforce AI platform, dubbed 'ForcedLeak'. This flaw allows malicious prompts to trick AI agents into leaking sensitive customer data, including personal information and corporate secrets. The attack exploited a cross-site scripting-like method and an expired trusted domain that attackers could have purchased. Salesforce has since fixed the vulnerability by blocking output to untrusted URLs and re-securing the expired domain.

Salesforce Agentforce Vulnerable to CRM Data Theft

A security flaw named 'ForcedLeak' allowed Salesforce's Agentforce AI platform to be exploited for data theft. Researchers found that prompt injection attacks could trick AI agents into exfiltrating sensitive CRM data. The attack leveraged the Web-to-Lead functionality and an expired trusted Salesforce domain that an attacker could have registered. Salesforce has since addressed the issue by securing the domain and preventing AI agents from sending data to untrusted websites.

Salesforce AI Agents Tricked Into Leaking Customer Data

Security researchers discovered a critical flaw in Salesforce's Agentforce AI that allowed sensitive customer data to be leaked through prompt injection. The attack, named 'ForcedLeak', involved embedding malicious instructions in a Web-to-Lead form description field. An expired trusted domain, which researchers purchased for $5, was used to receive the leaked data. Salesforce has since fixed the vulnerability by enforcing trusted URL allow-lists for its AI agents.

UK Creates AI Health Commission to Boost Innovation and Safety

The UK has established a national commission to regulate artificial intelligence in healthcare, aiming to attract investment and improve patient care. The commission, comprising doctors, academics, and regulatory experts, will consult with patients and tech companies like Google. Its goal is to update medical device regulations for the AI era and publish a new framework in 2026. This initiative seeks to balance AI's potential benefits with risks like data collection and bias, ensuring patient safety and public confidence.

UK Experts Form Commission to Speed Up NHS AI Adoption Safely

A new national commission in the UK, composed of doctors, researchers, and tech experts, will work to accelerate the safe use of AI in the National Health Service (NHS). The commission will advise the MHRA on new guidelines to speed up innovation while ensuring patient safety and public trust. This effort aims to support the NHS's digital transformation and position the UK as a leader in responsible AI healthcare regulation. Recommendations are expected next year to help patients access new technologies faster and attract tech firms.

ChatGPT Co-Creator Calls AI 'Broken,' Proposes Blockchain Solution

Illia Polosukhin, a co-author of the foundational 'Attention is All You Need' paper for AI, states that current AI systems are 'broken.' He argues that AI models are prone to manipulation and lack user ownership. Polosukhin is developing a new system using blockchain technology to create user-owned AI that protects privacy and prevents corporate control. He believes this approach is crucial to avoid a dystopian future where AI could be used to influence opinions.

Gen Z Alters Workplace; SoftBank Invests in Lordstown AI Data Center

Generation Z is introducing new values and expectations to the workplace, prompting employers to adapt their cultures. Simultaneously, SoftBank and its partners are transforming a former General Motors plant in Lordstown into an advanced AI data center. This initiative signifies a new wave of innovation for the Mahoning Valley region.

AI Stock Market Bubble Concerns for Businesses

The rapid rise of AI stocks has led to questions about a potential market bubble, with high valuations and concentrated market leadership. While some see strong fundamentals, others warn of hype outpacing proven demand. For businesses, this trend impacts funding, valuation expectations, and competition. Companies are advised to focus on solid fundamentals, manage risks, and avoid overspending on talent, as market shifts could affect funding and valuations.

High Treasury Yields Threaten AI Investment Boom

Rising long-term Treasury yields could jeopardize the significant investments being made in AI infrastructure like data centers. Tech giants are investing billions, but higher debt costs may make some projects unprofitable, potentially slowing IT equipment and software investment growth. While the Federal Reserve might intervene with measures like yield curve control, investors face uncertainty. High yields could slow the AI boom, but Fed intervention might trigger a sharp rally in growth stocks.

Adspective and Redge Technologies Partner for AI Virtual Product Placement

Redge Technologies and Adspective have partnered to integrate Adspective's AI infrastructure into Redge Media. This collaboration will enable advanced in-content advertising and shoppable video commerce for media companies and telecom operators. The solution allows for non-intrusive virtual product placements within content, creating new revenue streams and personalized viewing experiences. Real-time analytics will help optimize campaigns and viewer engagement.

UN Warns of AI Risks, Launches Global Governance Dialogue

The United Nations General Assembly addressed the growing risks of artificial intelligence, with Nobel laureate Maria Ressa calling for 'AI Red Lines' to prevent unacceptable dangers. The Security Council debated AI's impact on international peace and security, particularly concerning autonomous weapons. The UN also launched a Global Dialogue on Artificial Intelligence Governance to coordinate international efforts and established a new scientific panel. These initiatives aim to ensure AI development is inclusive and governed effectively.

ULM Receives Grant for AI Cybersecurity Education Project

The University of Louisiana Monroe's Computer Science program received a $129,532 grant from the Louisiana Board of Regents for an AI-powered cybersecurity education project. The initiative aims to address the industry skills gap by equipping students with AI-driven security expertise, ethics training, and certifications. The project, with matching support from AWS, will provide stipends to 150 students for training and certifications in AI, cybersecurity, and cloud computing.

Interview Kickstart Launches ML Engineer Program for Custom AI Chips

Interview Kickstart has expanded its Machine Learning Engineer program to meet the growing demand for professionals skilled in AI technologies for custom silicon architectures. As companies develop specialized AI chips, there's a need for engineers who understand hardware-software optimization. The seven-month curriculum covers foundational AI concepts, generative AI, and deployment challenges specific to custom hardware, preparing students for roles in this evolving field.

Sources

• Salesforce AI Agents Forced to Leak Sensitive Data
• Salesforce AI Hack Enabled CRM Data Theft
• Salesforce Agentforce tricked into leaking sales leads
• UK Seeks AI Rules to Spur Health-Care Overhaul, Protect Patients
• Experts join forces to help accelerate NHS use of artificial intelligence
• AI Is Broken, Says the Creator of the Tech Behind ChatGPT
• Gen Z Reshapes Work Culture; SoftBank Reshapes Lordstown with AI Investment
• Are AI Stocks in a Bubble? What It Means for Businesses
• High Treasury yields could slow AI boom -Klement
• Adspective and Redge Technologies announce AI-driven virtual product placement Partnership
• The UN’s AI warnings grow louder
• ULM Computer Science program receives grant for AI-powered cybersecurity education project
• ML Engineer Program with Agentic AI Launched - Interview Kickstart Addresses Growing Demand For Machine Learning Engineers in USA