Students are increasingly using tools such as EssayBot and Quillbot to help write college application essays, while some admissions offices also use AI to evaluate them. The number of AI-written essays remains unclear, but admissions teams are confronting drafts that lack the quirks of teenage writing, according to application editor Kate Gilbertson, a 2025 UC Berkeley graduate. Colleges generally discourage AI-written essays, yet the practice raises questions about fairness and how admissions officers should weigh context, intent and human judgment.
Canada plans to launch a nationwide AI literacy program in December, Federal AI Minister Evan Solomon said. The training is expected to take about 90 minutes and will be the third and final course in Ottawa’s Essentials for Educators program, which is expected to reach as many as one million students and 50,000 educators. It aims to help people use AI responsibly, assess generated content and recognize misinformation, bias and privacy concerns; Solomon has not specified the launch date or rollout details.
Debate over AI infrastructure is also sharpening. One analysis argues that market concerns center on heavy investment and uncertain revenue, not regulation, and that public resistance to data centers could slow expansion. That pressure might push developers such as OpenAI and Anthropic to focus less on training ever-larger models and more on inference and useful applications, where they can generate revenue. But existing models may already meet many business needs, and companies’ growing caution about AI means the strategy is not assured.
Ohio State University and Microsoft have announced a five-year AI partnership covering training, certifications, research data and a planned move to Azure. The effort will support AI work in healthcare, education and other industries, while giving students research, internship and job-placement opportunities. Ohio State associate vice president for research and innovation Dr. Bruce A. McPete and Microsoft AI for Healthcare general manager Julia Metcalf welcomed the collaboration.
Commonwealth Bank chief executive Matt Comyn says employers should train staff and give them time to adapt before using AI as a reason to cut jobs. He urged companies to explain early how roles may change, and said CBA’s Future Workforce Program will set out principles for practical support and finding new roles for affected employees. More than 27,600 CBA employees have participated in its internal AI learning series. Comyn warned that failing to adapt could weaken the country’s workforce.
Security reports highlight a gap between organizations’ use of AI agents and their ability to control them. Palo Alto Networks found that 99% of surveyed organizations use agents in development pipelines and 73% let them change code, pipeline settings or production environments. Yet they enforce fewer than three of eight measured security checks on average; 19% of serious application-security issues reach production. Two in five cannot reconstruct an agent’s actions, and only 36% can monitor AI behavior in production or 30% detect prompt injection, despite 97% feeling at least somewhat confident about security.
Detectify AB has introduced Agentic Tooling to help AI agents test potential vulnerabilities in live applications. The company says about 80% of findings from AI-based static application security testing tools are noise. Its Security Ledger records evidence for each test, and the tools work with existing agents through the Model Context Protocol or a command-line interface, using Detectify’s testing engines. CEO Rickard Carlsson said the aim is to provide proof an agent cannot fabricate.
A Microsoft service disruption that began November 12 affected Microsoft 365, including Outlook, Teams, Exchange Online, SharePoint and Copilot, disrupting access for organizations in more than 70 countries for over 24 hours. Microsoft said a problem tied to an earlier network-capacity upgrade resurfaced after a later network change, ruled out cyber activity and AI as causes, and confirmed restoration on November 14. Downdetector logged more than 7.4 million reports worldwide. Discussion also touched on Microsoft’s AI investment and October 2026 voluntary buyouts offered to about 60,000 employees, but there is no evidence the workforce changes caused the outage.
Researchers developed DECODE, a visual analytics system for finding and diagnosing failures in automated explanations of large language model representations. Because explanations are often generated and judged by LLMs, with limited human oversight and no clear ground truth, their reliability can be difficult to assess. DECODE organizes diagnosis into three stages, uses agreement among independently produced explanations as a reliability signal, and applies three-phase active seeding to choose cases for human review. Researchers tested it on 16,384 Sparse Autoencoder features and showed it can identify and trace failures at scale.
Key Takeaways
- AI-assisted college essays are complicating admissions as schools weigh fairness, authorship and the role of human judgment; the prevalence of AI-written submissions is unknown.
- Canada plans a roughly 90-minute national AI literacy program in December, focused on responsible use, misinformation, bias and privacy.
- An analysis argues that data-center opposition and uncertain returns could shift OpenAI and Anthropic toward inference and practical AI applications rather than ever-larger model training; success is not guaranteed.
- Ohio State and Microsoft launched a five-year partnership for AI training, research, Azure migration and student opportunities.
- CBA chief Matt Comyn urges employers to provide AI training and transition support before cutting jobs; more than 27,600 bank employees have joined its internal AI learning series.
- Palo Alto Networks found widespread AI-agent use but limited controls: organizations enforce fewer than three of eight security checks on average, and only 36% monitor AI behavior in production.
- Detectify’s Agentic Tooling records evidence for AI-agent security tests; the company says about 80% of findings from AI-based static testing tools are noise.
- A Microsoft 365 outage lasted more than 24 hours, affected organizations in over 70 countries and generated more than 7.4 million Downdetector reports; Microsoft said a network change triggered the issue and ruled out AI and cyber activity.
- DECODE helps diagnose failures in automated LLM explanations and was tested on 16,384 Sparse Autoencoder features.
AI use in college applications challenges admissions offices
Students are increasingly using AI tools such as EssayBot and Quillbot to write college application essays, though the article says it is not clear how many essays are AI-written. Colleges generally discourage the practice, but some also use AI to evaluate essays. Dr. Jane Smith, a college admissions officer, says schools must consider whether AI helped write an essay and understand its context and intent. The growing use of AI raises questions about fairness and the role of human judgment in admissions.
AI-written application essays leave admissions teams catching up
Students are using AI to help write college application essays, while colleges also use AI to evaluate them. The article says the shift is creating a scramble over who gets into the university of their choice. Kate Gilbertson, a 2025 UC Berkeley graduate who edits college applications full time, says she has recently received many drafts that lack the usual quirks of teenage writing.
Canada plans nationwide AI literacy training next month
Federal AI Minister Evan Solomon said Canada will launch its national AI literacy program for all Canadians in December. The training is expected to take about 90 minutes, though Solomon did not give a specific launch date or explain what the rollout will involve. It will be the third and final program in Ottawa’s broader Essentials for Educators course, which has been expected to reach up to one million students and 50,000 educators. The training aims to help people use AI more confidently and responsibly, assess AI-generated content, and spot misinformation, bias, and privacy concerns.
Data center slowdown could help AI developers focus on useful tools
The article argues that the current market crash reflects concerns about heavy AI investment and weak revenue prospects, rather than fears of regulation. Public opposition to data centers may slow their growth, but that could help AI developers shift from costly efforts to train ever larger models toward useful applications. Companies such as OpenAI and Anthropic are seeking large amounts of computing power to build new models, even though existing models may be capable enough for many business uses. The author says developers could instead use more computing power for inference, where revenue is generated, but notes that companies are becoming more cautious about AI use and this approach is not guaranteed to succeed.
Ohio State and Microsoft launch five-year AI partnership
Ohio State University and Microsoft announced a five-year partnership focused on AI training, research, and innovation. The work will include training, certifications, research data, and a planned move to Azure. The partnership aims to support AI solutions in healthcare, education, and other industries, and offer students opportunities in research, internships, and job placements. Ohio State associate vice president for research and innovation Dr. Bruce A. McPete and Microsoft AI for Healthcare general manager Julia Metcalf welcomed the collaboration.
CBA chief urges AI training before employers cut jobs
Commonwealth Bank chief executive Matt Comyn says employers should give staff AI training and time to adjust before cutting jobs. He urged companies to explain early how AI may change workers’ roles, saying this could help build skills and reduce opposition to the technology. CBA plans to publish principles through its Future Workforce Program, including practical support and help finding new roles for affected staff. More than 27,600 employees have taken part in the bank’s internal AI learning series. Comyn also warned that AI could weaken the country’s workforce if workers and businesses do not adapt.
Palo Alto Networks report finds gaps in AI agent security
A Palo Alto Networks report says organizations are struggling to track and secure AI agents in cloud systems. In the survey, 99% of respondents said they use agents in development pipelines, while 73% allow them to change code, pipeline settings or production environments. Organizations enforce fewer than three of eight measured security checks on average, and 19% of serious application security issues reach production. Two in five organizations cannot reconstruct an agent’s actions, while 99% have found ungoverned AI in the past year. Although 97% feel at least somewhat confident about securing AI, only 36% can monitor AI behavior in production and 30% can detect prompt injection.
Detectify launches tools to check AI security findings
Detectify AB introduced Agentic Tooling to help AI agents test possible vulnerabilities in live applications. The company says about 80% of findings from AI-based static application security testing tools are noise. Its Security Ledger records evidence behind each test so teams can separate verified threats from unsupported conclusions. The tools work with existing AI agents through the Model Context Protocol or a command-line interface and use Detectify’s testing engines. Chief Executive Rickard Carlsson said the goal is to give security teams proof that an agent cannot fabricate.
Microsoft outage highlights risks of relying on digital services
A Microsoft service disruption that began on November 12 affected Microsoft 365 services including Outlook, Teams, Exchange Online, SharePoint and CoPilot. Organizations in more than 70 countries faced problems accessing workplace tools, and the outage lasted more than 24 hours. Microsoft said an issue from an earlier network-capacity upgrade resurfaced after a later network change, and ruled out cyber activity and AI as causes. The company confirmed on November 14 that services had been restored; Downdetector recorded more than 7.4 million outage reports worldwide. Microsoft’s outage prompted discussion of its AI investment and workforce changes, including voluntary buyouts offered to about 60,000 employees in October 2026, but there is no evidence those changes caused the outage.
DECODE helps find failures in automated AI explanations
DECODE is a visual analytics system designed to help people find and diagnose failures in automated explanations of large language model representations. The process often uses LLMs to create and judge explanations, but limited human oversight and a lack of ground truth make it hard to check their reliability. DECODE divides diagnosis into three stages to help identify problems, from structural defects to mismatches in explanations. It uses agreement among independently produced explanations as a sign of reliability and a three-phase active seeding method to select cases for human review. The researchers tested the system on 16,384 Sparse Autoencoder features and showed that it can identify and trace failures at scale.
Sources
- As students rely on AI in college applications, admissions offices scramble to keep up
- As students rely on AI in college applications, admissions offices scramble to keep up
- Minister says Canada’s national AI literacy training will go live next month
- The Coming Data Center Slowdown Might be Good for AI Developers
- Ohio State and Microsoft partner on AI training and research | ETIH EdTech News
- CBA's Matt Comyn tells employers to provide AI training before layoffs
- Palo Alto Networks report finds organizations struggle to track and secure AI agents in the cloud
- Detectify launches Agentic Tooling to verify AI security findings
- Microsoft outage exposes risks of digital dependence as AI investment reshapes workforce
- DECODE: Diagnosing Interpretability Failures in Automated Interpretation Process
Comments
Please log in to post a comment.