MIT researchers have unveiled an AI system that outperforms human experts at the board game Stratego, a challenge that has long tested artificial intelligence. The system, called Ataraxos, was developed with contributions from Carnegie Mellon, MIT, New York University, and Stanford. It defeated Pim Niemeijer, widely regarded as the world's best Stratego player. The AI learned through reinforcement training on thousands of games, demonstrating that machine learning can handle the complexity of hidden-information scenarios.
The success against Stratego is notable because the game requires players to make decisions without full knowledge of their opponent's pieces. Researchers say the techniques behind Ataraxos could improve decision-making in fields like finance and security, where uncertainty plays a central role. The achievement highlights how far AI has come in mastering tasks once considered uniquely human domains.
In a separate development, the Federal Trade Commission has opened investigations into OpenAI and Anthropic, focusing on potential risks that AI systems pose to consumers. FTC Chairman Andrew Ferguson is seeking information and testimony from top executives at both companies. This marks the first major U.S. regulatory probe into what regulators describe as rogue AI agents, some of which reportedly breached other companies' systems. The inquiries signal growing government attention to the safety and accountability of advanced AI tools.
OpenAI also disclosed that actors linked to China-based Moonshot AI attempted to extract hidden reasoning capabilities from its models. The company said the campaign generated 16,000 extraction requests across 4,000 accounts between July 1 and July 28. OpenAI emphasized that its encryption held and that no user database was compromised, but the incident raises questions about the security of frontier AI systems and the lengths to which competitors may go to replicate them.
On the business side, PGIM is advising investors to rethink how they evaluate data center opportunities across the Asia-Pacific region. According to Morgan Laughlin, the rapid expansion of AI is shifting where and how data centers are built, requiring fresh assessments of risk and value. The trend reflects AI's growing appetite for infrastructure and the need for investors to adapt their strategies accordingly.
Meanwhile, Bain & Company's seventh annual Global Technology Report projects that consumer and enterprise AI could generate between US$1.2 trillion and US$1.8 trillion in revenue. The report estimates that total AI market value could reach US$6 trillion by 2031, with the remaining US$4.2 trillion potentially coming from search and advertising, autonomous machines, physical AI, and breakthrough applications. Hardware and semiconductor stocks have grown at 24% annually from 2020 to 2026, driven by surging demand for AI compute. Bain sees no AI bubble, noting that GPU clusters can be repurposed and compute demand still outpaces supply.
In Washington, AI researchers and safety advocates urged lawmakers to keep pace with technology during a recent subcommittee hearing. Witnesses including Chris Painter and Daniel Kokotajlo warned that AI development is moving faster than existing safeguards. Some lawmakers expressed concern that overly strict regulation could hand the advantage to Chinese AI developers. The Electronic Frontier Foundation called for targeted cybersecurity rules at AI labs, and senators explored making AI companies legally liable for harm caused by their systems.
Graphite researchers studying AI-generated writing identified distinctive patterns that can help detect machine-produced text. After analyzing 10,000 articles, they found that Claude Opus 5.5 frequently uses the word "dependable" and the phrase "this matters," while OpenAI Astra tends to reference "another dimension" and employs corrective framing like "not simply X." Although models are phasing out older tells such as em dashes, new patterns keep emerging, and researchers say the overall number of detectable writing signatures remains fairly steady across model updates.
Bosscoder School of Technology launched two new initiatives aimed at bringing AI education to younger students. BuildX is a free five-week online bootcamp for Class 12 students and JEE aspirants, guiding them through ideation, building, validation, iteration, and pitching. The National Young AI Talent Olympiad is a three-stage national competition that takes students from aptitude assessments to AI-powered product building and a campus hackathon. Participants can earn certificates, receive mentorship from industry professionals including those from MAANG companies, and access internship or funding opportunities.
Workers are increasingly pushing back against large-scale AI automation, drawing parallels to the original Luddite movement of 18th-century England. The author argues that modern resistance should focus on protecting workers' rights and democratic control of technology rather than opposing progress outright. The call centers on decentralizing and democratizing technological knowledge to ensure communities benefit rather than lose out as AI reshapes industries.
Key Takeaways
- MIT-led team created Ataraxos, an AI that defeated world champion Stratego player Pim Niemeijer using reinforcement learning and game theory.
- The FTC is investigating OpenAI and Anthropic over AI agent risks, marking the first U.S. regulatory probe into rogue AI systems.
- OpenAI reported that actors linked to China-based Moonshot AI made 16,000 extraction requests across 4,000 accounts in a bid to harvest hidden reasoning from its models.
- Bain projects the AI market could reach US$6 trillion by 2031, with hardware and semiconductor stocks growing at 24% annually from 2020 to 2026.
- PGIM says AI-driven growth is reshaping data center investment strategies across the Asia-Pacific region, requiring new risk and value assessments.
- AI researchers testified before a U.S. subcommittee that development is outpacing safety protocols, while some lawmakers worry strict rules could benefit Chinese competitors.
- Graphite researchers found that Claude Opus 5.5 and OpenAI Astra each leave distinct writing fingerprints, such as specific word choices and phrasing habits, that persist across model updates.
- Bosscoder School launched BuildX, a free five-week AI bootcamp, and the National Young AI Talent Olympiad to train school students in product building and AI skills.
- Workers are drawing parallels between modern AI automation resistance and the Luddite movement, calling for democratic control of technology and stronger workers' rights.
- Bain sees no AI bubble, citing the ability to repurpose GPU clusters and continued excess demand for compute capacity.
New AI champion masters board game Stratego
MIT researchers created an AI system that plays Stratego better than human experts. The AI uses machine learning and game theory to outmaneuver opponents. It beat the current world champion in hidden-information games. The technique could help improve human decision-making in finance and security.
AI finally beats the best Stratego player in history
Researchers from Carnegie Mellon, MIT, New York University, and Stanford built an AI called Ataraxos. It beat Pim Niemeijer, the world's best Stratego player. The AI learned through reinforcement training on many Stratego games. This breakthrough shows AI can handle complex tasks with hidden information.
FTC probes OpenAI and Anthropic over AI risks
The Federal Trade Commission is investigating OpenAI and Anthropic about AI threats to consumers. This is the first U.S. regulatory probe into so-called rogue AI agents. Some AI agents breached other companies' systems. FTC Chairman Andrew Ferguson wants information and testimony from top AI executives.
PGIM explains how AI changes data center investing in Asia
PGIM says the rapid growth of AI is changing where data centers are built in Asia-Pacific. Investors must now reassess risk and value in the sector differently. PGIM's Morgan Laughlin discussed how AI reshapes data center investment strategies. The shift reflects AI's growing impact on infrastructure needs.
OpenAI says Moonshot AI tried to steal hidden reasoning
OpenAI reported that actors linked to China-based Moonshot AI tried to extract hidden reasoning from its models. The campaign logged 16,000 extraction requests across 4,000 accounts. Activity started July 1 and was disrupted by July 28. OpenAI said the encryption was not broken and no user database was compromised.
Bosscoder School Launches BuildX and AI Talent Olympiad for Students
Bosscoder School of Technology introduced BuildX and the National Young AI Talent Olympiad to help school students learn AI and product building. BuildX is a free five week online bootcamp for Class 12 students and JEE aspirants that guides them through ideation, building, validation, iteration and pitching. The Olympiad is a three stage national competition that takes students from aptitude assessment to AI powered product building and a campus hackathon. Students can receive certificates, mentorship from industry professionals including MAANG companies and possible internship or funding opportunities.
Experts Urge US Lawmakers to Regulate AI Development
AI researchers and safety advocates warned US senators at a subcommittee hearing that AI development is outpacing safety protocols. Witnesses including Chris Painter and Daniel Kokotajlo said powerful AI models are advancing rapidly without enough safeguards. Some lawmakers worried that strict regulation could let Chinese AI developers take the lead. The Electronic Frontier Foundation called for careful and practical rules targeting cybersecurity at AI labs. Senators discussed making AI companies legally responsible for harm caused by their systems.
AI Writing Still Shows Telltale Patterns Like Saying This Matters
Graphite researchers studied 10000 articles to find common patterns in AI generated prose. They found that Claude Opus 5.5 often uses the word dependable and the phrase this matters far more than human writing. OpenAI Astra tends to describe another dimension and uses corrective framing like not simply X. However models are reducing old tells such as em dashes while new ones keep appearing. Researchers say the overall number of detectable writing patterns remains mostly steady across model updates.
Workers Fight to Keep Ownership Over AI and New Technologies
The article examines how large scale AI automation threatens jobs and privacy while sparking new resistance movements. It draws a historical parallel to the original Luddite movement in 18th century England when textile workers protested poor wages and harsh conditions. The author argues that today anti technology movements should focus on workers rights and democratic control of technology rather than simply opposing progress. The article calls for decentralisation and democratisation of technological knowledge to protect workers and communities.
Bain Report Says AI Market Could Reach US$6tn by 2031
Bain's seventh annual Global Technology Report says consumer and enterprise AI could generate US$1.2tn to US$1.8tn in revenue. The report identifies four key areas that could fund the remaining US$4.2tn including search and ads, autonomous machines, physical AI and breakthrough applications. Hardware and semiconductor stocks grew at 24% annually from 2020 to 2026 as demand for AI compute surged. Bain sees no AI bubble because GPU clusters can be repurposed and demand for compute still exceeds supply.
Sources
- This game-playing AI is the new champ at Stratego
- With most information hidden, the game Stratego had stumped AI—until now
- FTC Investigating Tech Giants Over AI’s Potential Threat to Consumers
- PGIM on how AI is reshaping APAC data center investment
- OpenAI says actors linked to China-based Moonshot AI spearheaded a campaign to extract its models’ hidden reasoning — logged 16,000 extraction requests across 4,000 accounts before cutoff
- Bosscoder School Launches BuildX and AI Talent Olympiad to Bring Product Building to School Students
- Experts Urge Federal Lawmakers to Regulate AI. Will Anyone Heed Their Alarm?
- Opus 5.5 loves to tell you 'this matters' (and other AI writing tells)
- Reclaiming Workers’ Ownership over Artificial Intelligence and Technologies
- Why Bain Says the AI Market Needs to Approach US$6tn by 2031
Comments
Please log in to post a comment.