Microsoft is reshaping its Copilot product into a single app that combines chat, document editing, coding, and AI agents in one place. The new design includes Home for productivity tasks, Code for building apps, and Autopilot for running recurring tasks automatically. CEO Satya Nadella described the update as bringing AI directly into the flow of work. Fewer than 7 percent of Microsoft 365 users currently hold Copilot licenses, so the company is looking for ways to grow adoption.
The relaunch puts Microsoft in a more direct contest with Anthropic's Claude and OpenAI's tools in the workplace. Companies enrolled in the Frontier program will get early access to the Home and Code sections soon. Autopilot will enter private preview at the end of September. Pricing is also shifting to include usage-based billing for certain features, which could change how businesses pay for these services.
In other AI news, Google introduced PageBreak, a system built by its Product Security team to find real security bugs in the company's own web apps. Revealed on September 24, the tool searches for vulnerabilities that could be exploited, offering a new approach to identifying and fixing security problems before they cause harm.
Meta is also making changes after researchers discovered a security flaw in its Muse AI assistant. The bug could have let attackers access private emails and files if a user manually approved a request from a malicious site. Meta initially rated the issue as severe but later downgraded it after reviewing the specific conditions needed to exploit it. The company also fixed a separate low-risk vulnerability on Mac computers that could have redirected audio recordings.
Across the Atlantic, the European Securities and Markets Authority is planning a 2027 review of how firms use AI and tokenization in finance. The focus is on products that directly affect customers, not just back-office work. Supervisors will look for risks like biased AI outputs and complex products that investors might not understand. The plan also expands cyber resilience checks to smaller firms and crypto-asset service providers.
In the Caribbean, the Caribbean Hotel Association partnered with Travel 2 Latam to help hotels adopt AI tools. Training workshops, webinars, and access to AI-powered tools will be offered to hotel managers. A recent survey found that 70 percent of hoteliers see AI as key to future success, yet only 20 percent have actually put these solutions into practice. The collaboration aims to close that gap.
On the security front, researchers developed DT4X+, an improved version of the DT4X algorithm for diagnosing faults in complex systems. The tool uses symbolic regression to create clearer rules that separate different types of system errors. Tests on several benchmark systems showed that DT4X+ delivers more informative results and higher performance than the original.
Separately, the ERRAND system is helping AI agents manage memory more efficiently. It treats knowledge updates as paid tasks that compete with main work, checking information only when resolving a doubt justifies the cost. In tests, ERRAND outperformed other methods by 10 percentage points while using only 11 percent of available steps.
Chinese companies are facing accusations from firms like Anthropic, OpenAI, and Google that they use a method called distillation to copy American AI technology. Many experts believe these claims are overblown. Charles O'Neill from Baseten said the narrative is not as true as people say, while Geoffrey Hinton compared the process to a teacher and student model relationship.
Akamai Technologies reported in September 2026 that it expects full year 2027 earnings of about $150 million to $300 million. Anthropic committed $11.6 billion over seven years with an option to increase total commitments to about $20 billion. The deal concentrates a large portion of future Akamai Cloud demand into one AI client, which could reshape capacity planning and revenue mix between content delivery and compute workloads.
AI-generated campaign ads attacking political candidates are growing in the 2026 midterms. One example shows New York City Mayor Zohran Mamdani and New York Governor Kathy Hochul in a fake conversation that paints them negatively. State laws require transparency about this practice, but candidates sometimes ignore them. Experts say better regulation is needed to address the increasing use of AI in political advertising.
Key Takeaways
- Microsoft is relaunching Copilot as a single app with Home, Code, and Autopilot sections to compete with Anthropic's Claude and OpenAI's tools in the workplace.
- Fewer than 7 percent of Microsoft 365 users currently hold Copilot licenses, and pricing is shifting to include usage-based billing for some features.
- Autopilot, a personal agent for running recurring tasks, will enter private preview at the end of September.
- Google built PageBreak, an AI system that finds real security bugs in its own web apps, revealed on September 24.
- Meta fixed a security flaw in its Muse AI assistant that could have let attackers access private emails and files through a malicious site request.
- The EU Securities and Markets Authority will review AI and tokenization use in finance starting in 2027, focusing on customer-facing products and risks like biased outputs.
- The Caribbean Hotel Association partnered with Travel 2 Latam to offer AI training and tools to hotels, where 70 percent of hoteliers see AI as key but only 20 percent have adopted it.
- DT4X+ improves fault diagnosis in complex systems using symbolic regression, delivering more informative results and higher performance on benchmark tests.
- ERRAND optimizes agent memory updates by treating them as paid tasks, outperforming other methods by 10 percentage points while using only 11 percent of available steps.
- Anthropic committed $11.6 billion over seven years to Akamai, with an option to increase total commitments to about $20 billion, reshaping Akamai's cloud demand outlook.
Microsoft updates Copilot to compete with Anthropic's Claude
Microsoft is updating its Copilot app for corporate workers to challenge Anthropic's Claude. The new app combines coding tools and productivity features into one place. Microsoft wants more Office 365 customers to pay for these advanced tools. The company is playing catch-up by adding coding features to the app. Companies in the Frontier program will get early access in coming weeks.
Microsoft relaunches Copilot as one app with coding and AI agents
Microsoft is relaunching Copilot as a single app that combines chat, document editing, coding, and AI agents. The app has three sections: Home for productivity tasks, Code for building apps, and Autopilot for running recurring tasks automatically. Fewer than 7 percent of Microsoft 365 users have Copilot licenses. Pricing is shifting to include usage-based billing for some features. The launch was confirmed by CEO Satya Nadella.
Microsoft bundles Copilot features to compete with OpenAI and Anthropic
Microsoft is bundling Copilot features into one product to compete with OpenAI and Anthropic in the workplace. The new Copilot includes Chat, Cowork, Code, and a personal agent called Autopilot. Microsoft CEO Satya Nadella described the update as bringing AI into the flow of work. Home and Code will roll out to Frontier program users soon. Autopilot will enter private preview at the end of September.
AI campaign ads grow in 2026 midterms despite disclosure laws
AI-generated campaign ads attacking political candidates are becoming common in the 2026 midterms. State laws require transparency about this practice, but candidates sometimes ignore them. Experts say it is too early to know how these ads will affect voters. More effective regulation is needed to address the growing use of AI in political advertising.
AI ads used in 2026 midterms lack disclosure despite laws
Campaign ads using AI to create misleading content about candidates appear in the 2026 midterms. One example shows New York City Mayor Zohran Mamdani and New York Governor Kathy Hochul in a fake conversation that paints them negatively. A voiceover describes Hochul as a political delusion. Experts stress the need for better rules on AI-generated political ads.
New DT4X+ Tool Improves Fault Diagnosis Accuracy
Researchers created DT4X+, an improved version of the DT4X algorithm for diagnosing faults in complex systems. The new tool uses symbolic regression to create clearer rules that separate different types of system errors. By fixing how the algorithm builds its training data, DT4X+ maintains better consistency and works more reliably on dynamic systems. Tests on several benchmark systems show that this enhanced version provides more informative results and higher performance.
ERRAND Algorithm Optimizes Agent Memory Updates
The ERRAND system manages agent memory by treating knowledge updates as paid tasks that compete with main work. It only checks information when the value of resolving a doubt justifies the cost of using actions. This approach ensures that agents stay accurate without wasting resources on unnecessary rechecks. In tests, ERRAND outperformed other methods by 10 percentage points while using only 11 percent of available steps.
Meta Upgrades Safety Alerts After Muse AI Bug
Meta Platforms is making safety warnings more visible in its Muse AI assistant after researchers found a security flaw. The bug could have allowed attackers to access private emails and files if a user manually approved a request from a malicious site. Meta initially rated the issue as severe but later downgraded it after reviewing the specific conditions needed to exploit it. The company also fixed a separate low-risk vulnerability on Mac computers that could have redirected audio recordings.
EU Regulators Plan 2027 Review of AI and Tokenization
The European Securities and Markets Authority will require national regulators to examine AI and tokenization use in finance starting in 2027. This new priority focuses on how firms use these technologies in products that directly affect customers rather than just in back-office work. Supervisors will check for risks like biased AI outputs and complex new products that investors might not understand. The plan also expands cyber resilience checks to include smaller firms and crypto-asset service providers.
Caribbean Hotels Partner to Adopt AI Tools
The Caribbean Hotel Association joined forces with Travel 2 Latam to help hotels in the region use artificial intelligence. Their partnership will provide training workshops, webinars, and access to AI-powered tools for hotel managers. A recent survey found that while 70 percent of hoteliers see AI as key to future success, only 20 percent have actually implemented these solutions. The collaboration aims to close this gap by offering specific resources and knowledge to improve the tourist experience.
Google Creates AI That Finds Its Own Security Flaws
Google built an AI system called PageBreak to find real security bugs in its own web apps. The Product Security team revealed this tool on September 24. The system hunts for vulnerabilities that could be exploited. It represents a new approach to finding and fixing security problems.
Are Chinese Companies Really Copying American AI?
Companies like Anthropic, OpenAI and Google have accused Chinese firms of using a method called distillation to copy their AI technology. Leaders are expected to discuss these claims as concerns about AI dangers grow. Many experts believe these accusations are overblown. Charles O'Neill from Baseten said the narrative is not as true as people say. Geoffrey Hinton compared it to a teacher and student model relationship.
Anthropic AI Deal Changes Akamai Investment Story
Akamai Technologies reported in September 2026 that it expects full year 2027 earnings of about $150 million to $300 million. Anthropic committed $11.6 billion over seven years with an option to increase total commitments to about $20 billion. The deal concentrates a large portion of future Akamai Cloud demand into one AI client. This could reshape capacity planning and revenue mix between content delivery and compute workloads.
Sources
- Microsoft packages business AI in single app as it tries to compete with Anthropic
- Microsoft is relaunching Copilot as a unified app combining coding and autonomous AI agents
- Microsoft Bundles Copilot Features to Challenge Anthropic and OpenAI in the Workplace
- AI ads are popular in the 2026 midterms, but state laws haven't stopped a lack of disclosure
- AI ads are popular in the 2026 midterms, but state laws haven’t stopped a lack of disclosure
- Discovery of fully efficient fault indicators along a data-based diagnosis process
- ERRAND: Budgeted Maintenance of Agent Memory
- META Stock Drops 3.4%: Meta Reportedly Moves To Strengthen Safety Alerts After Muse Security Issue Discovery
- EU Securities Watchdog Puts AI and Tokenization on Supervisors' Agenda for 2027
- Travel 2 Latam
- Google Built an AI That Hunts Its Own Security Bugs
- Is China Really Stealing A.I. From American Companies?
- How The Anthropic AI Deal At Akamai Technologies (AKAM) Has Changed Its Investment Story
Comments
Please log in to post a comment.