India's cybersecurity agency Cert-In has released a detailed set of AI security auditing guidelines that will affect any organization developing or deploying AI in the country, including foreign companies serving Indian users. The rules sort AI systems into four risk tiers, from low to critical, and set the pace and depth of audits accordingly. Organizations must report AI security incidents within six hours and keep security logs for at least 180 days. The guidelines arrive as CERT-In warns of AI-driven attacks, including a campaign called Crafty Falcon, targeting Indian banks, fintech firms and crypto exchanges.
AMD CEO Lisa Su painted a picture of AI becoming essential to everyday life and business. The company posted third quarter sales of $13.38 billion, up 45% year over year, with Data Center revenue hitting a record $8.5 billion, up 97%. Growth was driven by EPYC processors and Instinct MI450 AI accelerators. AMD guided for data center growth of 50% sequentially in the fourth quarter and expects server revenue to double by 2027.
Box is putting Nvidia's Nemotron embedding system to work in a new tool called Box Content Retrieval, aimed at helping AI agents pull relevant information more accurately. Nvidia claims the system can make queries four times faster while cutting errors by 20% and token use by up to 90%. The feature is set to roll out to Box Enterprise Advance customers early next year. Box also showed off an orchestration tool for agent workflows and threat detection for rogue agents.
In medical AI research, a new method called challenge-aware adaptive recruitment is helping systems find and fill factual gaps during reasoning. An external challenger inspects reasoning traces and triggers specialist model recruitment when gaps are found. On the MEDAGENTSBENCH hard subset of 862 questions, the approach improved accuracy across 8 of 9 datasets, lifting GPT-4.1-mini MedQA accuracy from 46% to 54%.
Pathkey.AI used its Chipforge platform to turn plain language design briefs into working hardware on a physical FPGA chip. An independent expert from Nanyang Technological University in Singapore validated the process. Two projects were completed, including a digital frequency monitor and an encryption design, and internal results on Nvidia's chip design benchmark scored above leading general-purpose AI models. Pathkey.AI is targeting paid projects across Asia Pacific and plans a broader Chipforge Studio release early next year.
Beacon Holdings landed a $50 million investment from CPP Investments to expand its AI software portfolio. The Toronto-based company runs 45 software businesses across recreation, utilities, education, government and manufacturing. Beacon uses a centralized AI operating system to share technology across its portfolio and plans to use the new funds for acquisitions and to grow its engineering team.
JPMorgan analysts say 2028 will be a key test for the AI data center buildout. Political pressure is mounting, with Democratic gubernatorial candidates in Texas, Iowa and Georgia pledging to restrict data centers after winning elections. Voters in several local initiatives also backed data center limits. Michael Cembalest from JPMorgan warned that the scale of construction could be curtailed, and the midterm elections have shifted the debate from whether data centers are at risk to how much their growth will be limited.
Key Takeaways
- Cert-In released 107-page AI security auditing guidelines classifying AI systems into four risk tiers: low, moderate, high and critical.
- Organizations must report AI security incidents within six hours and retain security logs for at least 180 days under new Indian rules.
- AMD reported third quarter sales of $13.38 billion, up 45% year over year, with Data Center revenue reaching a record $8.5 billion.
- Box integrated Nvidia's Nemotron embedding system into Box Content Retrieval, claiming queries could run four times faster with 90% fewer tokens.
- A new challenge-aware adaptive recruitment method improved GPT-4.1-mini MedQA accuracy from 46% to 54% on a hard subset of MEDAGENTSBENCH.
- Pathkey.AI's Chipforge platform turned written design briefs into working hardware on a physical FPGA chip, tested by an independent expert.
- Beacon Holdings secured $50 million from CPP Investments to expand its AI software portfolio across 45 businesses.
- JPMorgan warns that 2028 will test the AI data center buildout as political pressure grows to restrict data center construction.
- CERT-In flagged AI-driven attacks called Crafty Falcon targeting Indian banks, NBFCs, fintech firms and crypto exchanges.
- Carnegie Mellon University Africa is offering a free Certificate in Artificial Intelligence for Africa 2027 with six online courses.
Cert-In releases AI security audit rules with 4 risk tiers
Cert-In issued 107-page AI Security Auditing Guidelines on November 4. The rules apply to all organizations in India that develop or deploy AI, including foreign companies serving Indian users. AI systems are classified into four risk tiers: low, moderate, high and critical. Each tier determines how often and how deeply systems must be audited.
CERT-In requires six-hour reporting for AI security incidents
CERT-In issued new cybersecurity guidelines requiring organizations to report AI security incidents within six hours. AI systems will be sorted into four risk categories based on how serious a breach could be. Organizations must keep security logs for at least 180 days with synced timestamps. Audits must cover nine areas including data privacy, model risk and fairness.
AI-driven cyberattacks target India's financial sector
CERT-In warned of AI-powered cyberattacks called Crafty Falcon targeting Indian banks, NBFCs and fintech firms. Attackers use AI tools like the ARTEX family to run attack chains and bypass payment system APIs. Cryptocurrency exchanges and payment aggregators are also hit. CERT-In's new guidelines classify AI systems into four risk levels and require stronger audits.
Box uses Nvidia Nemotron for agent content retrieval
Box announced Box Content Retrieval at its developer conference, integrating Nvidia's Nemotron embedding system. The tool helps AI agents find relevant information faster and more accurately. Nvidia says it could make queries four times faster with 20% fewer errors and 90% fewer tokens. The feature will roll out to Box Enterprise Advance customers early next year.
Box partners with Nvidia for enterprise AI agent context
Box launched Box Content Retrieval at BoxWorks 2026, powered by Nvidia's Nemotron Retriever models. The system improves search accuracy by 20% and cuts token use by up to 90%. It also respects file permissions so agents cannot access unauthorized content. Box also introduced an orchestration tool for agent workflows and threat detection for rogue agents.
New AI system recruits knowledge during medical reasoning challenges
Researchers created a method called challenge-aware adaptive recruitment for medical AI. An external challenger inspects reasoning traces and finds factual gaps during problem solving. On the MEDAGENTSBENCH hard subset of 862 questions, this approach improved accuracy on 8 of 9 datasets. GPT-4.1-mini MedQA accuracy rose from 46% to 54% using challenger-triggered specialist recruitment.
AI debt, hedge funds and Africa credit rating news roundup
The World Economic Forum reported on growing financial concerns tied to AI. AI-related debt makes up nearly two-thirds of US venture-capital investment over the past year. Hedge fund assets are now triple their 2013 levels. Africa also launched its own credit rating agency to offer a new view of the continent's economies.
AMD CEO Lisa Su says AI will become essential in daily life
AMD CEO Lisa Su said AI and super intelligence will become essential to every part of life and business. AMD reported third quarter sales of $13.38 billion, up 45% year over year. Data Center revenue reached a record $8.5 billion, up 97%, driven by EPYC processors and Instinct MI450 AI accelerators. AMD guided for data center growth of 50% sequentially in the fourth quarter and server revenue doubling in 2027.
Fact check confirms AI helps filter email messages
The Roy Howard Community Journalism Center fact-checked claims about AI in email filtering. Major email providers are expanding AI use to catch spam, phishing and harmful messages. These systems can sometimes flag legitimate emails by mistake. Microsoft acknowledged more such cases, but sources did not confirm whether AI has caused more filtering errors.
Beacon Holdings gets $50 million to expand AI software portfolio
Beacon Holdings secured a $50 million investment from CPP Investments to grow its AI software portfolio. The Toronto-based company owns 45 software businesses serving recreation, utilities, education, government and manufacturing. Beacon uses a centralized AI operating system to share technology across its portfolio. The company plans to use the funds to acquire more software companies and expand its engineering team.
Pathkey.AI Chipforge Builds Working Hardware From Plain Language
Pathkey.AI used its Chipforge platform to turn written design briefs into working hardware on a physical FPGA chip. An independent expert from Nanyang Technological University in Singapore tested the platform without help from the company. Two projects were completed, including a digital frequency monitor and an encryption design. Internal results on NVIDIAs chip design benchmark scored above leading general-purpose AI models. The company is targeting paid projects across Asia Pacific and plans a broader Chipforge Studio release early next year.
Carnegie Mellon University Africa Offers Free AI Certification for 2027
Carnegie Mellon University Africa is accepting applications for a free Certificate in Artificial Intelligence for Africa 2027. The program includes six online courses covering topics like machine learning, deep learning and natural language processing. The certification is open to students from Africa who want to build skills in AI. Applications can be submitted through the university website. The program aims to help students prepare for careers in artificial intelligence.
Analysts Say 2028 Will Test the AI Data Center Buildout
JPMorgan analysts say 2028 will be a key test for the ongoing data center buildout tied to artificial intelligence. The concern grew after Democratic gubernatorial candidates won in states like Texas, Iowa and Georgia and pledged to restrict data centers. Voters in several local initiatives also supported data center restrictions or pauses. Michael Cembalest from JPMorgan warned that the scale of data center construction could be curtailed. The midterm elections shifted the focus from whether data centers are at risk to how much their growth will be limited.
Sources
- Cert-In issues AI security audit guidelines, sets 4-tier risk framework
- CERT-In sets six-hour reporting deadline for AI security incidents
- Mint Explainer | How AI-driven cyberattacks are targeting India's financial sector
- Box taps Nvidia’s Nemotron for new agent-first content retrieval system
- Box taps Nvidia's AI models to build a rich context layer for autonomous enterprise agents
- Challenge-Aware Knowledge Recruitment with Meta-Agent Supervision for Medical Multi-Agent Reasoning
- AI Debt Investment: Hedge Funds & Africa Credit Rating News
- AMD CEO Lisa Su sees AI becoming 'essential to every aspect of our lives'
- Fact-Check: AI plays a growing role in email filtering
- Beacon Holdings Secures $50 Million Investment from CPP Investments to Expand AI Software Portfolio
- Can Pathkey.AI (ASX:PKY) Chipforge Build Hardware Without Code?
- Free Carnegie Mellon University Africa AI Certification 2027: Apply for Six Online Artificial Intelligence Courses
- The real test for the artificial intelligence build-out is what happens in 2028, analysts say
Comments
Please log in to post a comment.