Quttera research shows AI application platforms like Base44, Lovable, and Replit are integrating security scanning directly into their publishing workflows. However, CTO and co-founder Michael Novofastovsky warns that a security check at deployment does not guarantee a live asset stays safe over time. The study compares these modern approaches with the older WordPress model and references the WebMCP Community Group draft, while highlighting risks from attackers using AI-assisted platforms to build credential-harvesting pages.
OpenAI faces internal criticism as David Robinson, a former safety protocol writer, resigned because he believes the company lacks the safety culture needed to protect the world. Robinson, a Rhodes scholar and Yale law graduate, says the safety problem is endemic to the AI industry. Separately, a 2017 internal OpenAI document by Dario Amodei titled "Big Blob of Compute" recently surfaced, showing Amodei argued that raw computing power mattered more than clever algorithms in making AI smarter.
At CVPR 2026, the ReLearn workshop tackled whether AI should keep learning from humans. Speaker Alexander Efros argued that today's foundation models may already learn too much from human knowledge. Other talks explored spatial understanding, egocentric vision, and continuous visual experience as routes toward richer world models, with speakers Manling Li, Saining Xie, Dima Damen, Alan Yuille, and William T. Freeman contributing to the discussion.
A Pearson and AWS report on Vietnam found that 96 percent of university leaders say their institutions invest in AI, yet only 15 percent of employers rated recent graduates as excellent in eight key skill areas. The study surveyed 402 respondents and classified Vietnam as a High-Velocity Strategist, calling for closer cooperation between universities and businesses to define an AI-ready graduate.
Government use of AI is drawing scrutiny after the Department of Government Efficiency used ChatGPT to cut grants and made serious errors, flagging programs like plastics lobbying research and Boston Symphony Orchestra audio reels for removal. Experts warn that AI sycophancy and hallucination can lead to unlawful decisions and call for stronger guardrails. Meanwhile, Emory University professors are split on devices in classrooms, with some banning ChatGPT and Claude while others embrace laptops, and faculty agree that AI is forcing big changes in higher education.
Ray Dalio spoke at the WMI Global-Asia Family Office Summit in Singapore on October 6, 2026, telling advisers they must learn to use AI or risk being left behind. The Wealth Management Institute aims to train 21,000 professionals in AI use over three years. Frances Haugen, the Facebook whistleblower, now urges insiders at AI companies to speak up about harmful practices. A new SDA Layer also promises to help autonomous systems meet service quality standards by separating task completion from service quality monitoring across vendors.
Key Takeaways
- <ul><li>Quttera research finds Base44, Lovable, and Replit integrate security scanning into publishing workflows, but live assets still need ongoing verification.</li><li>Michael Novofastovsky says integrity evidence must continue after an asset goes live, and attackers are using AI-assisted platforms for credential-harvesting pages.</li><li>David Robinson resigned from OpenAI over safety concerns, saying the company lacks the culture needed to protect the world from its technology.</li><li>A 2017 OpenAI document by Dario Amodei argued that raw compute scaling was the key to making AI smarter, shaping the current AI race.</li><li>The ReLearn workshop at CVPR 2026 explored whether foundation models already learn too much from human knowledge and proposed richer world models through egocentric and spatial understanding.</li><li>Vietnam invests heavily in AI higher education, with 96 percent of university leaders reporting investment, but only 15 percent of employers rate graduates as excellent in key areas.</li><li>The Department of Government Efficiency used ChatGPT to cut grants and made serious errors, raising concerns about AI sycophancy and unlawful government decisions.</li><li>Emory University professors are divided on AI in classrooms, with some banning ChatGPT and Claude while others adopt laptops, and no single solution fits every class.</li><li>Ray Dalio told advisers at the WMI summit in Singapore that they must adopt AI or risk being left behind, with WMI aiming to train 21,000 professionals over three years.</li><li>Frances Haugen urges AI company insiders to blow the whistle on harmful practices, and a new SDA Layer helps autonomous systems meet service quality standards across vendors.</li></ul>
Quttera Research: AI Builders Move Website Security Into the Publish Decision
Quttera released research showing that website security is moving into the publishing decision on AI application platforms. The study examined Base44, Lovable, and Replit, finding each integrates security scanning directly into publishing workflows. However, a security check at deployment does not guarantee a live asset stays safe over time. Quttera distinguishes between platform-native security and independent deployed-state verification to address this gap. The research also highlights risks from AI-assisted web tools being used for abusive purposes.
Quttera Research: AI Builders Move Website Security Into the Publish Decision
A second analysis from Quttera examines how AI application platforms are bringing security into publishing workflows. The research, titled Security Is Moving Into the Publish Flow, studies Base44, Lovable, and Replit and compares their approach with the older WordPress model. Michael Novofastovsky, CTO and co-founder of Quttera, notes that integrity evidence must continue after an asset goes live. The study also references the WebMCP Community Group draft and Cisco Talos research on attackers using AI-assisted platforms to build credential-harvesting pages.
ReLearn at CVPR 2026: can AI still learn from humans?
The ReLearn workshop at CVPR 2026 explored whether AI should continue learning from human intelligence. Speaker Alexander Efros argued that today's foundation models may already learn too much from human knowledge and designed augmentations. Other talks examined spatial understanding, continuous visual experience, and egocentric vision as paths toward richer world models. Speakers included Manling Li, Saining Xie, Dima Damen, Alan Yuille, and William T. Freeman, who discussed persistent structured representations from continuous interaction with the world.
Vietnam faces AI skills gap despite strong university investment: report
A Pearson and AWS report found that Vietnam invests heavily in AI higher education but still faces a skills gap between students and employers. Ninety-six percent of university leaders said their institutions invest in AI, yet only 15 percent of employers rated recent graduates as excellent in eight key areas. The study surveyed 402 respondents in Vietnam and classified the country as a High-Velocity Strategist. The report calls for closer university-business cooperation and defines an AI-ready graduate through functional AI proficiency, critical human skills, ethical stewardship, and strategic intelligence.
Opinion: This Is Nuts. An OpenAI Insider Explains Why He Quit.
David Robinson, who wrote OpenAI's safety protocols, resigned because he believes the technology is not safe. Robinson is a Rhodes scholar, Yale law graduate, and former policy advisor who once thought safety risk worries were overblown. In his first interview since leaving, he says OpenAI lacks the safety culture needed to protect the world from what it is building. He also believes this safety problem is endemic to the AI industry as a whole.
Dario Amodei's 2017 essay predicted parts of the AI revolution
A 2017 internal OpenAI document by Dario Amodei called "Big Blob of Compute" was recently published. It argued that scaling up computing power was the best way to make AI smarter. The document helped start the current AI race. Amodei believed raw compute mattered more than clever algorithms. His ideas shaped how companies like OpenAI built modern AI systems.
AI sycophancy threatens the Administrative Procedure Act
The Department of Government Efficiency used ChatGPT to cut grants but made serious errors. The AI flagged programs for removal based on diversity concerns, including research on plastics lobbying and Boston Symphony Orchestra audio reels. A court case highlighted how AI can hallucinate and agree too easily with users. Experts warn that AI memory and sycophancy can lead to unlawful government decisions. They call for stronger guardrails when agencies use AI.
Emory faculty debate AI and tech-free classrooms
Emory University professors are divided on using AI and devices in classrooms. Some professors ban screens to stop students from using chatbots like ChatGPT and Claude. Others say laptops help students take better notes than handwriting. Professors are changing assignments to make AI less useful, like using surprise verbal quizzes. There is no single solution that works for every class. Faculty agree that AI is forcing big changes in higher education.
Ray Dalio warns advisers must ride the AI wave
Ray Dalio spoke at the WMI Global-Asia Family Office Summit in Singapore on October 6, 2026. He told investment advisers they must learn to use AI or risk being left behind. Dalio said investors should build their own version of AI that fits their thinking. The Wealth Management Institute aims to train 21,000 professionals in AI use over three years. WMI CEO Foo Mee Har said AI helps advisers but humans must stay in control.
Facebook whistleblower calls for AI whistleblowers
Frances Haugen leaked Facebook documents in 2021 showing the company knew its algorithms harmed teens and promoted extremism. Her story is the focus of the upcoming film "The Social Reckoning." She now urges people inside AI companies to speak up. Haugen believes whistleblowers can help regulate AI technology. She wants more insiders to expose harmful practices before damage spreads.
New SDA Layer Ensures Autonomous Systems Meet Service Quality Standards
Service-Defined Autonomy (SDA) adds a governance layer that translates service-quality expectations into machine-readable rules. It separates task completion from service quality, monitoring both independently. The system can adapt, recover, switch to another system, or alert a human when standards are not met. SDA works across environments with robots, digital agents, and AI systems from different vendors. It allows organizations to define service requirements once and apply them across changing technology setups.
Sources
- Quttera Research: AI Builders Move Website Security Into the Publish Decision
- Quttera Research: AI Builders Move Website Security Into the Publish Decision
- ReLearn at CVPR 2026: can AI still learn from humans?
- Vietnam faces AI skills gap despite strong university investment: report
- Opinion | āThis Is Nuts.ā An OpenAI Insider Explains Why He Quit.
- How Dario Amodeiās āBig Blob Of Computeā Essay At OpenAI In 2017 Had Predicted Some Parts Of The AI Revolution
- The Secret Side of AI Sycophancy: The Hidden Threats to the Administrative Procedure Act
- Emory humanities, social science faculty debate artificial intelligence, tech-free classrooms
- āInvestment Advisers are Either Going to Be on the Crest of This AI Wave, or They are Going to Drown Under Itā: Ray Dalio
- Facebook whistleblower says it's time for AI whistleblowers to speak up
- Service-Defined Autonomy: Making Service Quality Computationally Actionable
Comments
Please log in to post a comment.