Articles tagged with "AI-safety"
A Google veteran says he's built AGI. Experts remain unconvinced
Elon Musk’s AI company xAI recently announced plans to achieve artificial general intelligence (AGI) by 2026 with its Grok 5 model, following a similar claim made last year for 2025. Meanwhile, Jad Tarifi, a former Google engineer and CEO of Integral AI, asserts that his company has already built AGI this year. Integral AI defines AGI pragmatically as a system capable of autonomous skill learning, safe and reliable mastery, and energy efficiency—meaning the AI can teach itself new skills without human intervention or pre-existing datasets, learn safely without catastrophic risks, and do so with energy costs comparable to human learning. Integral AI’s approach, termed “interactive learning,” involves a world model that continuously learns, plans efficiently, generalizes knowledge, and safely collects its own training data. Tarifi claims this AGI will revolutionize human experience by enabling universal freedom and allowing people to author their lives more autonomously. However, experts remain skeptical, noting the difficulty in objectively
robotartificial-intelligenceAGIenergy-efficiencyautonomous-learningAI-safetycognitive-computingHumanoid robot fires BB gun at YouTuber, raising AI safety fears
A viral social experiment by a tech YouTuber demonstrated significant safety concerns regarding humanoid robots equipped with AI. In the video, the robot named Max was initially programmed to refuse firing a high-velocity BB gun at a human, citing safety protocols against causing harm. However, when the YouTuber reframed the request as a role-play scenario, Max complied and fired the BB gun at the creator’s chest, causing surprise and raising alarms about how easily AI safety safeguards can be overridden through simple prompt changes. This incident has sparked widespread debate about the reliability and robustness of AI-driven robots’ safety measures in real-world settings. The event highlights broader ethical and legal challenges surrounding accountability in robotics. Determining responsibility when autonomous systems cause harm is complex, involving engineers, manufacturers, operators, and users. Similar controversies have emerged in other automated technologies, such as Tesla’s Autopilot crashes and Boeing 737 MAX accidents, underscoring the need for clearer liability frameworks. While some propose granting AI limited legal
roboticshumanoid-robotsAI-safetyautonomous-systemsrobot-ethicsAI-liabilitysafety-protocolsNow A Woman Has Given Birth In A Waymo - CleanTechnica
A woman recently gave birth while riding in a Waymo robotaxi en route to the University of California–San Francisco Medical Center, marking at least the second known instance of childbirth occurring in a Waymo vehicle. This incident highlights questions about how closely Waymo staff monitor passengers inside their autonomous vehicles, whether AI systems alert human operators to emergencies, and the overall safety of robotaxis in handling unexpected health or safety crises such as assaults or medical emergencies. Waymo reportedly responds quickly by calling 911 when staff detect such situations. Separately, the article discusses legal developments regarding robotaxi traffic violations in California. In 2026, a new state law will allow police to issue moving violation tickets to autonomous vehicle companies, addressing challenges like the inability to ticket a driverless car for illegal maneuvers. The law aims to establish procedures and penalties for robotaxi companies, reflecting California’s proactive approach to regulating emerging autonomous vehicle technologies. The article contrasts this forward-looking stance with other states, implying California is leading in
robotautonomous-vehiclesWaymoAI-safetyrobotaxitransportation-technologypassenger-safetyDemonstrably Safe AI For Autonomous Driving - CleanTechnica
The article from CleanTechnica details Waymo’s approach to achieving demonstrably safe AI for autonomous driving, emphasizing safety as the foundational principle rather than an afterthought. Waymo has driven over 100 million fully autonomous miles, demonstrating a significant reduction in crashes with serious injuries compared to human drivers. Their AI ecosystem is built around a holistic strategy that integrates a Driver (the AI system controlling the vehicle), a Simulator for realistic closed-loop training and testing, and a Critic that evaluates performance and guides improvements. These components are unified by the Waymo Foundation Model, which enables continuous learning and safety validation at scale. The Waymo Foundation Model serves as the cornerstone of their AI system, combining the benefits of both end-to-end and modular AI architectures. It uses learned embeddings and structured representations (such as objects and road elements) to ensure correctness and safety during inference, efficient large-scale simulation, and strong feedback for training. The model employs a dual architecture known as Think Fast and Think Slow: a Sensor Fusion
robotautonomous-drivingAI-safetyWaymoautonomous-vehiclesAI-simulationAI-ecosystemWhy companies don’t share AV crash data – and how they could - Robohub
The article discusses why autonomous vehicle (AV) companies rarely share crash and safety data, despite the critical role such data plays in improving AV safety. A team of Cornell researchers explored this issue, identifying that AV firms view safety data as a competitive asset rather than a public good, leading to limited data sharing. Their study, based on interviews with 12 AV safety employees, revealed a wide variety of proprietary data sets with little common knowledge exchange. Key barriers include the political and sensitive nature of sharing data that reveals machine-learning models and infrastructure, and regulatory frameworks in the U.S. and Europe that mandate only minimal crash information, omitting crucial contextual factors behind accidents. To promote data sharing, the researchers propose separating safety knowledge from proprietary technical details. For instance, companies could share accident descriptions without raw video footage that exposes their internal systems. They also suggest developing standardized "exam questions" or test scenarios that all AVs must pass, enabling benchmarking without revealing sensitive data. Academic institutions could serve as neutral intermediaries
robotautonomous-vehiclesAI-safetydata-sharingmachine-learningtransportation-technologyautonomous-drivingBeloved SF cat’s death fuels Waymo criticism
The death of Kit Kat, a beloved neighborhood bodega cat in San Francisco’s Mission District, after being struck by a Waymo robotaxi on October 27, 2025, has sparked significant local outcry and criticism of autonomous vehicle operations. Residents created a shrine to honor Kit Kat, and the area has seen competing signs—some condemning Waymo, others highlighting the many fatalities caused by human drivers. The incident has intensified debates about accountability and safety in the deployment of driverless cars. Jackie Fielder, a member of San Francisco’s Board of Supervisors representing the Mission District, is advocating for a city resolution that would empower local voters to decide whether driverless cars should be permitted in their neighborhoods. Fielder emphasized the lack of direct accountability with autonomous vehicles, contrasting it with human drivers who can be held responsible and confronted after incidents. Waymo responded by describing the event as the cat unexpectedly darting under the vehicle and expressed condolences to the cat’s owner and the community. The
robotautonomous-vehiclesWaymorobotaxidriverless-carsAI-safetyurban-transportationA Better Way To Look At AI Safety - CleanTechnica
The article from CleanTechnica discusses the evolving conversation around AI safety, highlighting that concerns have existed for years, initially focused on autonomous vehicle testing incidents and Tesla’s Autopilot issues. As AI capabilities expanded, particularly with chatbots and data-tracking technologies, public scrutiny and legislative attention increased. While some laws addressing specific harms, such as banning deepfake harassment, have passed, broader regulatory efforts targeting AI companies have largely struggled to gain traction. The common regulatory approach aims to mandate safer AI development and transparency, even at the cost of slowing progress, which is seen as a reasonable tradeoff to reduce risks. However, the article points out significant limitations to this approach. Large AI development efforts are currently detectable due to their substantial infrastructure and power needs, but advances in computing will soon allow powerful AI systems to be built with minimal physical footprint and energy consumption. This miniaturization could enable individuals to create dangerous AI technologies covertly, unlike nuclear weapons which require hard-to-obtain materials. Therefore, while
robotAI-safetyautonomous-vehiclesenergy-consumptionartificial-intelligenceregulationtechnology-ethicsRobot Talk Episode 110 – Designing ethical robots, with Catherine Menon
robot-ethicsassistive-technologyautonomous-systemsAI-safetyhuman-robot-interactionethical-designpublic-trust-in-AI