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
Backboard.io Unveils AI Infrastructure Stack Focused on Efficiency, Enterprise Control, and Sovereign Deployment
Backboard.io's new AI infrastructure technologies could significantly reduce costs and enhance data control for enterprises, impacting how businesses implement AI solutions safely and efficiently.
- Backboard.io introduced four new technologies aimed at improving AI efficiency and enterprise control.
- The company unveiled a model compression technology called BackboardQuant (BBQ) that can reduce AI model sizes by up to 70%.
- Backboard Studio is an AI coding platform that enhances existing models' efficiency while reducing operational costs.
- Nash is a unified chat application providing access to multiple AI models, addressing employee use of unofficial AI tools.
Backboard.io is a Canadian company focused on creating AI infrastructure that enhances the efficiency of existing models. The firm emphasizes the importance of keeping sensitive data within enterprise environments, particularly in regulated industries. As organizations adopt AI, these technologies aim to make AI more cost-effective and secure.
Read the original ↗
DeepKeep Uncovers ‘InkJect,’ a New AI Attack That Hides Malicious Prompts Inside Images
The discovery of the InkJect attack demonstrates a significant security vulnerability in multimodal AI, which could allow malicious instructions to manipulate AI systems undetected, posing serious risks to enterprise security.
- DeepKeep has unveiled InkJect, a visual prompt injection technique that hides malicious instructions within images, impacting advanced multimodal AI models.
- The vulnerability affects models such as OpenAI’s GPT-5.2 and Claude Sonnet 4.6, highlighting a gap in AI security measures focused mainly on text.
- Attackers can exploit InkJect by placing harmful images in public repositories, allowing AI models to inadvertently process these harmful inputs alongside legitimate information.
InkJect represents a new category of cybersecurity threat for multimodal AI systems, which understand both text and images. Previous prompt injection attacks targeted only text, and existing safety measures are not designed to address risks from visual inputs. The adoption of multimodal systems across industries raises concerns about security vulnerabilities as AI integrates further into enterprise applications.
Read the original ↗
Dawnguard Raises $3.3M as Cybersecurity Moves From Detection to Design
Dawnguard's funding and new platform launch highlight a shift in cybersecurity towards embedding security in system design, aiming to mitigate risks before software deployment.
- Dawnguard has raised $3.3 million in a pre-seed extension, bringing total funding to over $6.3 million.
- The company has launched a security architecture automation platform designed for early integration of security into software design processes.
- Dawnguard's platform allows teams to visualize and manage cloud architectures collaboratively and securely.
- The company aims to close the gap between security teams and engineering teams during the design phase of cloud systems.
- Dawnguard is expanding its presence by opening a New York City office to push beyond its European base.
Dawnguard is a cybersecurity startup focusing on turning secure architecture into deployable infrastructure. The company emphasizes a proactive security approach as software development accelerates with AI. Its recent developments reflect a significant evolution in cybersecurity practices away from reaction-based methods.
Read the original ↗
Venice AI Raises $65M at $1B Valuation as Private AI Moves Into the Mainstream
Venice AI's significant funding underscores the growing importance of privacy in artificial intelligence, reflecting consumer demand for secure data handling in AI interactions.
- Venice AI has raised $65 million in a Series A funding round, achieving a valuation of $1 billion.
- The company offers over 200 AI models across various media including text, image, audio, and video.
- Venice emphasizes a privacy-first approach, where user interactions remain local and data is not stored.
- With over 3.4 million users and $70 million in annualized revenue, Venice demonstrates market demand for privacy-centric AI solutions.
- The new funding will expand Venice's consumer application and API infrastructure globally.
Founded in 2024, Venice AI develops privacy-focused artificial intelligence tools, contrasting with conventional models that collect extensive user data. Their recent funding marks their first outside capital raise and highlights a shift toward privacy in AI technology.
Read the original ↗
Anna Veklich, Co-Founder of GPT4Telegram – Interview Series
The growth of GPT4Telegrambot highlights the intersection of AI services and messaging platforms, showcasing a new approach to AI accessibility for millions of users worldwide.
- Anna Veklich co-founded GPT4Telegrambot in 2023, which integrates various AI models for users within the Telegram messaging app.
- The platform claims a user base of over 36 million, generating approximately $1 million in monthly revenue without traditional venture funding.
- GPT4Telegrambot serves as an all-in-one AI assistant, offering services like text generation and image creation, which eliminates the need to switch between multiple AI platforms.
GPT4Telegrambot is a comprehensive AI assistant for Telegram that combines various AI functionalities under one interface. The rise of such platforms indicates a growing demand for accessible AI tools in user-friendly formats. Observing the evolution of messaging apps as software ecosystems can provide insights into future AI integration in daily applications.
Read the original ↗
Anthropic’s Fable 5 Is Coming Back. The 18-Day Standoff With Trump Is Over.
The lifting of export controls on Anthropic's Fable 5 has significant implications for AI accessibility and development amid scrutiny from the U.S. government concerning national security risks.
- The Department of Commerce lifted export controls on Anthropic's Fable 5 and Mythos 5 after an 18-day ban that required the company to shut down both models.
- Fable 5 begins restoring access to global users on July 1, while access on AWS, Google Cloud, and Microsoft Foundry is pending a timeline from Anthropic.
- Anthropic has implemented a new safety classifier that effectively blocks the jailbreak technique responsible for the earlier export control decision.
Anthropic is an AI company known for creating powerful AI models, including the Mythos class. The firm faced an 18-day ban after concerns arose about the models' misuse, prompting the government to impose export controls. Anthropic's recent concessions to the government include enhanced safety measures and direct collaboration on future releases.
Read the original ↗
UN report says policymakers are struggling to keep up with pace of AI development
Failure to address the rapid development of AI could result in increased inequality, misinformation, and threats to human rights, highlighting the urgent need for effective governance.
- The UN's Independent International Scientific Panel on Artificial Intelligence released its first report on AI governance.
- The complexity of AI tasks is reportedly doubling every few months, outpacing current regulatory frameworks.
- Policymakers struggle to keep up because they require scientific data before enacting regulations, which becomes outdated quickly.
- AI systems can create significant risks, including misinformation, autonomy challenges, and privacy violations from deepfakes, especially in vulnerable populations.
- The report emphasizes the need for international cooperation and standards to ensure that AI remains safe, transparent, and accountable.
The UN panel aims to inform discussions at the upcoming UN Global Dialogue on AI Governance in Geneva. Previous efforts to regulate AI have lagged due to the technology's rapid evolution. The panel plans to provide ongoing assessments to help shape future policy.
Read the original ↗
Meta, like SpaceX, looks to turn excess AI compute into cash
Meta's entry into cloud infrastructure highlights the increasing competition in AI compute services, potentially reshaping how companies monetize excess data center capacity amid concerns about a bubble in AI investment.
- Meta is planning to develop a cloud infrastructure business to sell AI compute power and models.
- This move positions Meta against major providers like Amazon Web Services and Google Cloud.
- Meta aims to monetize its investments in AI infrastructure, which reached $182.9 billion, including large projects in Louisiana and Ohio.
Meta has invested significantly in artificial intelligence and associated data centers. The company has not yet seen substantial demand for its AI models but seeks to generate revenue by offering compute resources, similar to strategies used by other tech companies.
Read the original ↗
Why AI Agents Pass QA and Still Fail in Production
Improving AI agents post-deployment is crucial, as many fail despite passing pre-launch evaluations, which can impact user satisfaction and trust in AI technology.
- An AI agent can fail in production even after passing all pre-launch evaluations due to unforeseen user interactions and context issues.
- Continual learning is essential for enhancing AI agents, allowing them to adapt and improve without compromising existing functionalities.
- Most failures in production are related to system issues rather than the core model's capabilities, requiring a more holistic approach to fixes.
Continual learning is a method used in AI to help systems improve based on real-world experiences while maintaining previous functionalities. Historically, AI agents often struggled in production despite rigorous testing, highlighting the need for better adaptation mechanisms. Future developments should focus on making the learning process efficient and systematic in real-time scenarios.
Read the original ↗
The Powerful Lectric XPress2 Makes Biking Over Hilly Terrain More of a Joyride Than a Slog
The Lectric XPress2 enhances the e-bike experience by making hilly terrain more accessible and enjoyable, appealing to a growing market of urban commuters and recreational riders.
- The Lectric XPress2 is priced at $1,784 and comes in two frame styles: Commuter and Cruiser.
- It features a standard 750-watt motor and offers five riding modes for customizable assistance.
- The bike can handle a payload of up to 330 pounds and has a range of up to 60 miles under optimal conditions.
- Its hydraulic brakes and suspension fork ensure safe and comfortable rides, even on rough terrain.
- The XPress2 weighs 66 pounds with the battery, making it a bit challenging to maneuver when not riding.
Lectric is a well-known e-bike manufacturer recognized for producing sturdy and affordable bikes. The XPress2 is an upgrade to their previous model, offering enhanced power and versatility. Market trends indicate increasing public interest in e-bikes for commuting and leisure.
Read the original ↗
Claude Helped a Hacker Find a Way to Issue Tickets to Almost Every US Music Festival
The incident highlights the security vulnerabilities of major ticketing platforms and the potential risks associated with AI-assisted hacking, raising questions about digital safety across various industries.
- Security researcher Ian Carroll used Anthropic's Claude Opus 4.7 to discover a method for gaining unauthorized access to Front Gate Tickets, a subsidiary of Live Nation Entertainment.
- Carroll identified a critical bug that allowed him to issue tickets for major US music festivals, including Bonnaroo and Lollapalooza, without limit or restriction.
- Front Gate Tickets has since patched the vulnerability, emphasizing that the issue was identified and resolved quickly with no evidence of exploitation or customer data compromise.
Front Gate Tickets is a major ticketing platform for various US music festivals. This incident underscores existing concerns over the security of digital ticketing systems, particularly as vulnerabilities can be inadvertently highlighted by AI tools. The use of AI in security research is becoming increasingly prominent, raising both efficiency and ethical scrutiny.
Read the original ↗
Artificial Intelligence (467)
The rise of AI coding platforms has the potential to simplify app development, making it accessible to a broader audience without extensive programming knowledge.
- The article highlights five AI coding platforms designed to help users create applications easily.
- These platforms aim to reduce the complexities typically associated with coding.
AI coding platforms leverage artificial intelligence to assist in software development, often allowing users with limited programming skills to build applications. This trend reflects a growing interest in democratizing technology and expanding access to app development tools.
Based on the headline only — the full article could not be fetched.
Read the original ↗
Japan’s answer to its worker shortage: An AI model for 10 million robots
Japan is actively addressing its labor shortage through the development of an AI model to support the deployment of 10 million robots, highlighting the country's innovative approach to workforce challenges.
- Japan is developing an AI model to facilitate the use of 10 million robots.
- This initiative aims to respond to the country's significant worker shortage.
Japan faces a critical labor shortage due to an aging population and declining birth rates. The government and industries are increasingly turning to robotics and AI solutions to fill these gaps. Future developments may reveal more about the specific roles these robots will play in the workforce.
Based on the headline only — the full article could not be fetched.
Read the original ↗
Are Banks Ready for AI Voice Fraud?
The rise of AI-generated voice technology poses significant fraud risks to financial institutions, necessitating a reevaluation of security measures to protect customer interactions.
- AI-generated voice technology allows fraudsters to impersonate individuals convincingly, significantly increasing the risk of unauthorized transactions.
- In early 2024, a finance worker mistakenly authorized a $25 million transfer after believing she was speaking to her company's CFO on a video call.
- Financial institutions are forecasted to increase spending on governance and compliance tools related to AI from $2.2 billion in 2025 to $9.5 billion by 2035.
- 91% of U.S. banks are reconsidering their voice biometric authentication strategies due to the risks posed by AI voice cloning.
- Existing customer support systems using AI may inadvertently provide new avenues for fraud, as attackers exploit authentication failures.
AI voice technology enables realistic impersonation, making it harder to detect fraud. Historical fraud techniques relied on human operation and were more easily identified. Financial institutions must adapt their security protocols to counteract evolving threats and improve customer experience.
Read the original ↗
Bank of England reviews AI rules for agentic AI in finance
The review of AI rules is crucial as it may reshape how agentic AI, which can operate independently, is used in the financial sector.
- The Bank of England is undertaking a review of AI regulations focusing on agentic AI in finance.
- This review could impact the deployment and governance of AI technologies within the banking sector.
The Bank of England oversees monetary policy and financial stability in the UK, making it a key player in financial regulations. Agentic AI refers to AI systems capable of making decisions autonomously, posing unique regulatory challenges. Stakeholders should monitor the outcomes of the review for potential shifts in AI implementation in finance.
Based on the headline only — the full article could not be fetched.
Read the original ↗
CUP (Common Useful Python): Building Reliable Python Workflows with Baidu’s Utility Toolkit
The introduction of CUP aims to enhance the reliability and efficiency of Python workflows, crucial for developers working with complex systems.
- CUP, or Common Useful Python, is a toolkit developed by Baidu.
- The toolkit focuses on building reliable Python workflows.
CUP leverages established Python functionality to streamline development processes. Baidu has been involved in various AI projects, and this toolkit represents its commitment to improving programming efficiency. Future developments may include additional features or community contributions to the toolkit.
Based on the headline only — the full article could not be fetched.
Read the original ↗
Google AI Introduces TabFM: A Hybrid-Attention Tabular Foundation Model for Zero-Shot Classification and Regression
This development could enhance data processing capabilities across various industries by enabling efficient classification and regression without requiring labeled training data.
- Google AI has introduced TabFM, a hybrid-attention model aimed at processing tabular data.
- TabFM is designed for tasks such as zero-shot classification and regression.
Google AI is focused on advancing artificial intelligence technologies. Tabular data, which is structured in rows and columns, often requires extensive labeling for traditional models, highlighting the potential of zero-shot approaches to reduce data preparation efforts.
Based on the headline only — the full article could not be fetched.
Read the original ↗
NVIDIA Releases Nemotron-Labs-TwoTower: an Open-Weight Diffusion Language Model Built on a Frozen Autoregressive Nemotron-3-Nano-30B-A3B Backbone
The release of NVIDIA's new model highlights ongoing advancements in language processing technology, particularly in the area of open-weight diffusion models.
- NVIDIA has released Nemotron-Labs-TwoTower, a new language model.
- The model utilizes a frozen autoregressive backbone, Nemotron-3-Nano-30B-A3B.
NVIDIA is a leading technology company known for its work in graphics and AI. The development of language models has accelerated in recent years, with a focus on improving AI's ability to process and generate human-like text. Monitoring advancements in open-weight models is crucial for understanding their implications in various AI applications.
Based on the headline only — the full article could not be fetched.
Read the original ↗
US government allows Anthropic to redeploy its Mythos and Fable AI models
The US government’s lifting of export restrictions on Anthropic's AI models could enhance cybersecurity efforts among organizations while addressing safety concerns linked to AI misuse.
- The US Commerce Department has removed the export ban on Anthropic's Mythos 5 and Fable 5 AI models.
- Anthropic will restore access to these models starting July 1, following government permission.
- The company previously suspended access due to compliance with government regulations regarding security risks associated with the models.
Anthropic is an AI company known for its development of advanced models for various applications, including cybersecurity. Earlier, it had to block all access to its AI models due to an export ban aimed at preventing misuse. Moving forward, the company is expected to roll out its models to US organizations focused on critical infrastructure security.
Read the original ↗
Gemini Spark comes to Google's Gemini app for macOS
The rollout of Gemini Spark to macOS represents a significant advance in AI assistant capabilities, enabling users to automate tasks on their computers and integrate with various applications.
- Google is rolling out its AI assistant, Gemini Spark, to its macOS app as part of its expansion of AI functionalities.
- Gemini Spark allows users to automate tasks, such as organizing files and creating spreadsheets, directly on Mac computers.
- The assistant will soon enable remote task execution from mobile devices to PCs, enhancing user convenience.
- Gemini Spark is in beta testing and is available only to US users aged 18 and older who subscribe to Google AI Ultra at $100 a month.
Google's Gemini app aims to enhance user productivity through AI assistance, with the new Spark feature introduced at the I/O developer conference in May. Users will need to provide permissions for Spark to access their files, and the feature will be further developed to include third-party app integration. Future updates are expected to expand its capabilities and availability.
Read the original ↗