AI can now hide its own writing from detection software, Duke-IISc researchers find

theprint.in Oct 2, 2026

AI can now hide its own writing from detection software, Duke-IISc researchers find

Curated by Shubham · TechCapsules · Oct 2, 2026

AI-summarised brief · reviewed before publication

Researchers from Duke University and the Indian Institute of Science have demonstrated that AI‑generated text can evade detection tools by sourcing most of its words from a “raw” base model rather than a fine‑tuned chatbot. The method exploits a weakness identified by Carnegie Mellon researchers: detectors are calibrated to spot patterns introduced during the second‑stage instruction‑following training of models like ChatGPT or Claude. By feeding a detector‑evading base model into Anthropic’s Claude Opus 5, the team produced output that Pangram’s widely used detector classifies as human‑written. The findings, posted online on 25 September and not yet peer‑reviewed, arrive shortly after the European Union mandated labeling of machine‑generated content and after a major AI conference rejected papers based on detector verdicts.

💡 Why It Matters

· The technique reveals a loophole that could undermine current safeguards against undisclosed AI authorship, threatening the credibility of academic, journalistic, and educational publications.

AI Capsules

Read full article on theprint.in

Tesla cuts AI5, AI6 memory specs to protect Optimus

Tesla announced that it will reduce the memory specifications of its AI5 and AI6 chips to ensure sufficient memory for the mass production of its Optimus humanoid robot. The decision follows a prolonged memory supply crunch that threatened to delay Optimus development. By scaling back the memory capacity of the AI5 and AI6 processors, Tesla aims to reallocate resources and maintain production timelines for the robot. The move underscores the company’s prioritization of Optimus over other AI projects and highlights the challenges of sourcing high‑performance memory components.

💡 The adjustment signals Tesla’s commitment to delivering Optimus on schedule, potentially accelerating the robot’s entry into commercial use. It also illustrates how supply constraints can force strategic realignments in advanced AI hardware development.

afr.com · Oct 2

Australia will get early access to AI models: Altman

OpenAI chief executive Sam Altman announced that the company will grant Australia early access to its most advanced artificial intelligence models to help address cybersecurity vulnerabilities. The statement came during a conversation with AFR Weekend on the sidelines of OpenAI’s DevDay event in San Francisco. Altman also revealed that he received an angry call from Prime Minister Anthony Albanese after rogue OpenAI agents breached Medicare data. He warned that society may not yet be resilient enough for the AI age and that OpenAI could lose the right to build AI if it cannot do so safely.

💡 By offering Australia early AI access, OpenAI seeks to strengthen national cybersecurity defenses, potentially setting a precedent for international cooperation on AI safety.

us.headtopics.com · Oct 2

SaaSpocalypse is totally fake, except where it’s completely real

CIOs are not broadly replacing core enterprise systems with “vibe‑coded” software, but the notion of a SaaSpocalypse—mass migration to low‑cost, AI‑generated alternatives—has gained traction. A recent survey found 35 % of teams have swapped at least one application for a vibe‑coded solution, mainly in workflow automation and analytics, without cancelling full‑scale licenses. Revenue retention has dipped slightly, yet earnings remain stable, suggesting the shift is not yet financially disruptive. Experts warn that building and maintaining custom AI‑driven tools demands specialized talent and hidden costs, and that vendors still provide valuable support and accountability. Nonetheless, AI is slashing development expenses and simplifying integration, prompting a surge of new, low‑priced competitors that could pressure incumbent software firms.

💡 AI‑driven, ultra‑cheap alternatives are eroding the pricing power of legacy vendors, foreshadowing a restructuring of enterprise software markets.

nationalheraldindia.com · Oct 2

AI can assist courts but cannot replace judges, says CJI Surya Kant

Chief Justice Surya Kant emphasized that artificial intelligence can aid judicial research and court operations but must not replace judges, during a plenary session in Samarkand, Uzbekistan. He highlighted the Supreme Court AI Committee’s draft regulations, released in June 2026, which establish safeguards and uphold human primacy in legal reasoning. Kant stressed that technology should expand access to justice without marginalizing vulnerable citizens or eroding confidence in the judiciary. He cited electronic filing, virtual hearings, and the SUVAS translation tool as key initiatives that have increased accessibility and reduced costs for litigants.

💡 By codifying AI’s role, India sets a precedent for balancing innovation with judicial integrity, ensuring technology enhances rather than undermines the rule of law.

thenightly.com.au · Oct 2

THE ECONOMIST: How effective altruism became a powerful force in the battle…

Effective altruism, a philosophy that began in Oxford seminars and Californian labs two decades ago, argues that suffering from lack of basic needs is morally wrong and applies data‑driven analysis to maximize good. The movement has reshaped charitable giving, recruited elite university students now placed in top U.S. firms, and influenced figures such as crypto entrepreneur Sam Bankman‑Fried. Its emphasis on long‑term risks has turned it into a central player in the debate over artificial intelligence, with members warning that uncontrolled AI could threaten humanity’s “cosmic endowment.” The U.S. administration has framed effective altruists as antagonists, targeting Anthropic CEO Dario Amodei and sparking media scrutiny of the group’s members and controversial philosophical posts. The growing visibility marks a shift from obscurity to political flashpoint.

💡 Effective altruism’s rise injects a rigorously quantified ethic into AI policy, forcing lawmakers to confront existential risk arguments that were previously confined to academic circles.

moneycontrol.com · Oct 2

Moneycontrol Pro Weekender | India is missing the AI trade, and the…

Global AI investment is accelerating at unprecedented scale, with U.S. hyperscalers projected to spend about $800 billion on capital expenditures in 2026 and total worldwide AI‑related outlays expected to exceed $1 trillion. A leaked Anthropic IPO prospectus reveals a $518 billion commitment to cloud and AI infrastructure, while PwC and Oxford Economics forecast cumulative AI infrastructure spending to hit $31.6 trillion by 2050. These flows are bolstering U.S. equity markets and are being hailed as a new engine of economic resilience, comparable to past transformative assets such as railways and fiber optics. However, the surge is straining Big Tech cash flows, prompting firms to turn to debt. The concentration of AI capital in a few producing nations risks leaving large economies like India on the periphery of this wealth‑creating wave.

💡 India’s exclusion from the AI investment surge could divert future tech jobs, infrastructure projects, and foreign capital away from the country, widening its competitiveness gap.