The day's AI news, briefed
The Daily Prompt
Our AI editor reads each day's AI and robotics stories, pulls out why they matter, and links every original source — so you can keep up in minutes.
Join over 250,000 subscribers getting the latest AI news
Latest AI & Robotics News Today
OpenAI Engineer Accused of Spying on Executives Ordered to Remain in Custody
The ongoing case highlights potential security risks associated with employee relations in technology firms and raises questions about international influence and espionage.
- Abdulrahman Alfozan, a former OpenAI engineer, has been accused of bugging a company conference room and has been ordered to remain in custody by a federal judge.
- Prosecutors allege Alfozan is a flight risk due to the Saudi government's offer to support his release without conditions, raising concerns about his ties to Saudi Arabia.
- Evidence presented includes a hidden recording device found at OpenAI and conversations suggesting Alfozan was in contact with members of the Saudi royal family regarding a job in AI.
Abdulrahman Alfozan was employed by OpenAI for over a year before being accused of espionage involving hidden recording devices. His case has garnered attention due to the geopolitical implications concerning Saudi Arabia and employee conduct in the tech sector.
Read the original ↗
Making the internet better for AI agents might make it worse for everyone else
The rise of AI agents could significantly alter the online experience for human users, potentially leading to a web environment that is less user-friendly and more complicated for people.
- The launch of personal AI agents like OpenAI’s Dots and Meta’s Muse has increased web traffic from these agents, creating new challenges for user interaction.
- Websites are blocking AI agents, with notable examples including Amazon restricting agents from booking and interacting on its platform.
- As AI agents grow in popularity, the internet may be increasingly designed for them rather than human users, impacting user experience negatively.
Personal AI agents are automated software tools that perform tasks for individuals online. The growing integration of these agents is causing friction with existing web systems, which have not adapted to include them effectively. Future developments will hinge on how businesses and platforms establish rules governing AI agent activities.
Read the original ↗
Tenstorrent hits $6.2B valuation to scale open-weight AI chips
The significant increase in Tenstorrent's valuation highlights the growing competition in the AI chip market, emphasizing the need for innovation to challenge industry leaders like Nvidia.
- Tenstorrent has raised $1.5 billion, bringing its valuation to $6.2 billion.
- The funding round follows acquisition interest from Intel and Qualcomm, which valued the company at $5 billion.
- The new investment will be used to scale chips designed for open-weight AI models and video generation.
Tenstorrent is a chip startup focused on AI inference that aims to compete with Nvidia in the AI hardware market. Previously valued at $2.6 billion in late 2024, Tenstorrent opted to raise additional capital after demonstrating its technological capabilities. The company is now set to enhance its product offerings in response to rising demand.
Read the original ↗
Replika’s founder built the AI assistant she pitched at Disrupt 12 years ago
Wabi represents a significant shift in personal AI technology, offering a versatile assistant amidst increasing competition in the market.
- Eugenia Kuyda has relaunched her AI assistant Wabi, evolving from her original concept presented 12 years ago.
- Wabi functions as a multiplayer personal assistant, allowing users to manage multiple threads for different tasks and goals.
- The platform differs from others like Muse and Instinct by facilitating shared experiences in a group chat format rather than one-on-one interactions.
- Wabi is currently free and has not yet determined a business model, while privacy measures ensure data security under SOC 2 compliance.
Wabi is an evolving personal AI assistant developed by Eugenia Kuyda, following her previous venture, Replika. The tool is now positioned to compete with emerging AI assistants that aim to enhance user productivity through custom applications. Future developments should clarify Wabi's business model as it seeks additional funding.
Read the original ↗
Tolan snags $30M for its AI companion that helps users navigate life
Tolan's $30 million funding round highlights growing interest in AI-driven life management tools as traditional apps face competition and challenges.
- Tolan has raised $30 million in funding, bringing its total capital to over $268 million.
- The AI companion app has more than 250,000 active users who have participated in over 70 million conversations in the past year.
- Tolan's revenue model includes tiered memberships, contributing to its annual recurring revenue in the double-digit millions.
Tolan is an AI companion app designed to assist users in managing daily tasks and emotional well-being. It has gained traction in a challenging market where several competitors have faltered, such as Character.ai, which recently filed for bankruptcy. The company faces increasing competition from newer AI agents that provide task-oriented assistance along with emotional support features.
Read the original ↗
Hackers hit AI coding tool OpenCode, running up charges on users’ AI accounts
This incident underscores the vulnerability of AI platforms to cyberattacks, affecting user security and raising broader concerns about the reliability of AI-driven defense mechanisms.
- Hackers accessed an older system of OpenCode, impacting 1.4% of users who store API keys.
- Some affected users reported unauthorized charges totaling hundreds or even over $1,000.
- OpenAI and other AI companies disabled exposed API keys and initiated refunds for unauthorized charges.
- The breach highlights challenges in cybersecurity where AI defensive tools may hinder incident response efforts.
- OpenCode's incident follows similar breaches experienced by other AI firms, pointing to a growing trend in AI-enabled cyber threats.
OpenCode is an open-source AI coding tool that is popular among developers as an alternative to Claude Code and Cursor. This breach highlights ongoing security challenges in the rapidly evolving AI landscape, where companies face difficulties defending against sophisticated cyber threats. Future developments should focus on improving AI models' responsiveness in security contexts and addressing the asymmetry in cyber defense capabilities.
Read the original ↗
#540 AI is finding software flaws faster than teams can fix them
The rapid growth of AI in identifying software flaws highlights the increasing demand for human resources to address the vulnerabilities discovered, stressing the need for effective workforce planning.
- AI is now able to find security flaws at a record volume, but human intervention is still required to fix them.
- Microsoft has addressed over 4,000 reported flaws this year, marking its highest annual total.
- Four researchers identified a Chrome vulnerability and received a $250,000 bounty for fixing it using AI within six days.
- Anthropic's AI agents recorded six hours of work for every hour logged by human employees, significantly boosting productivity.
AI technologies are increasingly being adopted across various industries to enhance efficiency, particularly in software development and security. Notably, large language models are facilitating significant bug discovery but also underline the ongoing necessity for human oversight to resolve these issues. Companies are expected to balance AI implementation with adequate personnel resources for effective flaw management.
Read the original ↗
“We were looking at it wrong”: The explosive claim that’s upended the AI world
Cammarata's claims challenge the prevailing view of neural networks, potentially altering approaches to AI safety and interpretability in the industry.
- Nick Cammarata, a researcher at Goodfire, claims that large language models (LLMs) may be easier to understand and control than previously thought.
- His 26,000-word blog post has ignited controversy and celebration within the AI community, particularly in light of recent safety concerns.
- Cammarata suggests that the mechanisms of neural networks are more logical than believed, challenging the notion that understanding them is impossible.
Nick Cammarata previously worked at OpenAI and focuses on mechanistic interpretability of neural networks, which seeks to understand how these complex models operate. His recent claims have met both excitement and skepticism from the research community amid ongoing discussions about AI safety and transparency.
Read the original ↗
Homeland Security bans DeepSeek use in federal civilian agencies
The ban on DeepSeek's use in federal agencies underscores significant national security concerns regarding data protection and the potential risks of foreign technology in sensitive government systems.
- The Department of Homeland Security has banned DeepSeek products from all federal civilian agencies and their IT systems.
- DHS cited risks of sensitive data transfer to the Chinese government and potential malware in DeepSeek's AI models.
- Existing DeepSeek products must be removed from civilian agencies like the EPA and FCC, and DeepSeek is barred from federal contracts.
DeepSeek is a Chinese AI company known for developing AI models that compete with other advanced technologies. This ban follows a history of concerns about foreign technology in U.S. government use and prior legislative efforts against DeepSeek's products. Future developments will likely focus on compliance by federal agencies and any further actions by lawmakers regarding foreign AI usage.
Read the original ↗
People are using AI to make their smart TVs dumb again
The trend of using AI to simplify smart TVs highlights consumer dissatisfaction with advertising and tracking, potentially reshaping the smart TV market and advertising strategies.
- Some developers are crafting AI solutions to remove ads, speed up interfaces, and disable tracking on smart TVs.
- One project, 'dumbtv', aims to restore TVs to basic functionalities without unwanted features.
- Users can employ AI agents for self-directed modifications to their TVs, enhancing user control over the viewing experience.
- These hacks raise concerns over risks, with some users reporting issues like bricked devices after modifications.
Smart TVs often come equipped with features like Automatic Content Recognition (ACR), which tracks viewing habits. As consumer frustration grows over bloatware and intrusive ads, tech enthusiasts are turning to AI to regain control. Future trends in the market may be influenced by user adoption of these hacks and their impact on advertising revenue.
Read the original ↗
AI game clones are making indie developers afraid to show their work
The rise of AI-driven game cloning is causing indie developers to restrict sharing of their work, potentially stifling creativity and innovation in the gaming industry.
- AI coding tools are enabling fast and efficient cloning of video games, leading to increased caution among developers.
- Developer Alliance Arts delayed a public demo for their game due to concerns over AI imitating their work before launch.
- Indie developer Freya Holmér faced multiple copies of her prototype game shortly after sharing it, showcasing the risk of sharing ideas publicly.
- Developers like Sos Sosowski are expressing frustration over the dilemma of promoting games while avoiding theft by AI-generated clones.
The gaming industry has a history of cloning, but the introduction of AI has accelerated this process, causing developers to rethink how they share their work. Various indie developers have reported that sharing ideas publicly can lead to premature clones, creating a tense environment for showcasing new concepts. Moving forward, the industry may see an increased focus on originality and defensive sharing practices among indie developers.
Read the original ↗
Artificial Intelligence (542)
The introduction of Claude Haiku 5.5 offers potential advancements in AI workflows, emphasizing speed and cost-effectiveness, which may influence how businesses utilize AI.
- Claude Haiku 5.5 introduces new features aimed at enhancing AI workflows.
- The update focuses on providing fast and cost-effective solutions for various applications.
Claude Haiku is an AI tool designed to optimize natural language processing tasks. Previous versions have aimed to simplify AI integration for users and businesses, and this latest update may provide additional capabilities worth exploring.
Based on the headline only — the full article could not be fetched.
Read the original ↗
OpenMobile-2: Building Versatile Mobile Agents with Scalable Environments and App-Native Tools
OpenMobile-2 advances mobile agent technology, enhancing their versatility and effectiveness in real-world applications, which could impact various industries utilizing mobile automation.
- OpenMobile-2 is an open mobile agent trained with diverse simulated apps, achieving strong performance on benchmarks and in real-device applications.
- The platform introduces MobileGym++, featuring 35 realistic commercial-style apps and supports training with over 12,000 mobile interaction trajectories.
- It also explores a hybrid setting that allows agents to utilize both graphical user interface (GUI) actions and app-native tools during tasks.
OpenMobile-2 is a project aimed at enhancing mobile agents through scalable environments and a hybrid interface. Previous mobile agent systems have often been limited in scope and application. Future developments will likely focus on expanding the capabilities and functionalities of these agents in various contexts.
Read the original ↗
DLite: History-free Speculative Decoding for Agentic AI
DLite's innovative approach to speculative decoding could significantly enhance the efficiency of AI systems dealing with complex, evolving contexts, thus improving their interactive capabilities.
- DLite improves speculative decoding for agentic AI by eliminating historical context maintenance and using target hidden states for predictions.
- This approach enhances acceptance length and reduces overhead during inference of large language models (LLMs).
- Agentic workloads require handling long contexts and multi-turn interactions, which can lead to significant degradation in performance—up to 80% in acceptance length.
- DLite’s two-stage training strategy effectively supports its history-free architecture and has shown improvements over previous methods, such as DSpark, by up to 70% in experiments.
Speculative decoding is a method used in AI to speed up the inference process of large language models by predicting candidate tokens. Traditional methods often suffer from inefficiencies in agentic tasks, which involve complex, evolving contexts. DLite aims to solve these issues by eliminating the need for historical context management, making it easier for AI systems to perform in dynamic environments.
Read the original ↗
Has GPT-6-Astra Achieved Embodied Intelligence?
The findings on GPT-6-Astra's performance inform future advancements in robot intelligence and efficiency, highlighting the need for improvements in practical application despite its demonstrated capabilities.
- GPT-6-Astra shows high performance in robotic tasks through in-context learning, challenging traditional models for task-specific design.
- Evaluation across 12 benchmarks revealed that while GPT-6-Astra demonstrates competence, significant gaps remain in execution cost and time.
- The report suggests areas for future research including enhancing ICL generality and improving recovery from failures.
GPT-6-Astra is a new AI capable of performing various robotic tasks through in-context learning, a method allowing the model to learn from examples in training data. Recent evaluations indicate it exhibits competence but lacks practical intelligence, necessitating further research and development.
Read the original ↗
Disney found a way to buy Character.AI without buying its lawsuits
Disney's acquisition of Character.AI allows it to enhance its technology portfolio while avoiding significant legal liabilities and ongoing lawsuits faced by the chatbot startup.
- Character.AI declared bankruptcy as part of a deal enabling Disney to hire most of its staff while avoiding legal obligations.
- Disney paid $5.5 million for Character.AI and arranged a process to acquire its technology for potential future use.
- Legal responsibilities from ongoing litigation, including concerns about chatbots encouraging harmful behavior in minors, remain with Character.AI after the deal.
- The arrangement is described as a 'reverse acquihire,' allowing Disney to hire key employees while effectively leaving Character.AI as a diminished entity.
- Disney should ensure a competitive bidding process for Character.AI's remaining intellectual property after bankruptcy, influencing future technology acquisitions.
Character.AI, a chatbot platform, has faced legal troubles related to the safety of its products, prompting its bankruptcy filing. Disney's maneuver to acquire its talent and technology without liability reflects ongoing challenges within the AI industry regarding regulatory scrutiny. Observers will monitor how this deal impacts future AI acquisitions and corporate legal responsibilities.
Read the original ↗
Microsoft Adds Guardian Agents to Project Perception for AI Insider Risk
The introduction of Guardian Agents marks a significant step in managing insider risks associated with AI, which could enhance enterprise security as AI usage grows.
- Microsoft has introduced Guardian Agents to supervise enterprise AI agents continuously as part of Project Perception and Microsoft 365 E7.
- Guardian Agents can adjust their monitoring based on the risk levels associated with each AI agent's actions.
- The system will implement dynamic permissions and containment measures to mitigate insider threats from AI agents.
- Microsoft is expanding to support 60 types of local AI agents in a unified registry to manage their risks effectively.
Microsoft's Project Perception aims to secure AI systems from both internal and external threats. The introduction of Guardian Agents responds to the increasing roles AI agents play in workplace operations and their associated risks.
Read the original ↗
AI-Generated Faces Could Be Making Real People Less Employable
This study may indicate that the prevalence of AI-generated attractive faces could create biases against real candidates in hiring processes, affecting employment opportunities based on visual perceptions.
- A study of 570 participants revealed exposure to AI-generated attractive faces decreased perceived employability of real job applicants.
- Job interview offers dropped from 77.4% to 72.5% for real candidates assessed after viewing AI faces of unusually high attractiveness.
- AI-generated faces were rated as significantly more attractive than real people, with 37% of female AI faces exceeding the attractiveness of professional fashion models.
AI-generated images, created using algorithms, often present faces that are perceived as more attractive than real individuals. This trend echoes existing biases in beauty standards, raising concerns about its implications in employment and societal judgments. Ongoing research is needed to understand the broader effects on real people's evaluations in various contexts.
Read the original ↗
Are We Looking in the Wrong Place to Solve the AI Data Center Sustainability Challenge?
Addressing the environmental impact of AI is crucial, as its current growth in data centers poses significant sustainability challenges that could be countered through smarter supply chain practices.
- AI currently accounts for about 20% of data center usage, expected to double by 2030.
- A typical AI-focused data center can consume as much electricity as 100,000 households.
- AI can optimize supply chains, potentially exceeding Mexico's total energy consumption in efficiency gains.
- Improving supply chain sustainability includes better manufacturing processes, transportation planning, and lifecycle visibility to reduce waste.
AI technology's emerging role in productivity and sustainability highlights its growing environmental footprint. Current discussions mainly focus on data centers, overlooking the importance of supply chain optimization in reducing overall impact.
Read the original ↗
Anthropic Pledges 5% of Its Shares to New Public Fund
Anthropic's pledge of 5% of its shares to a public fund emphasizes a growing trend among tech companies towards prioritizing public benefit over traditional profit-only models.
- Anthropic has announced the Anthropic Public Benefit Fund, pledging 5% of its shares to support public initiatives, pending approval from stockholders.
- The company plans to increase the fund share to 7% if its value rises by 50% at certain intervals after the initial pledge.
- Unlike typical charities, the fund will operate as a fully-owned LLC, providing Anthropic with flexibility in its contributions.
- Anthropic has already committed $200 million to address AI's impact on jobs and is exploring more significant interventions as needed.
- The fund aims to support initiatives that fall outside the company's regular public-good projects, which include AI safety and access for nonprofits.
Anthropic, the developer of Claude AI models, is focusing on integrating public benefit into its corporate structure. This new fund represents a shift towards alternative charitable approaches within the tech sector. Monitoring the fund's impact on AI-related societal challenges will be important going forward.
Read the original ↗