Amazon is expanding the tools around Bedrock AgentCore. LoanPro, whose platform serves more than 450 lenders and manages loans reaching about 31% of U.S. households, built a multi-agent assistant with AWS’s Generative AI Innovation Center and AllCloud to summarize accounts, find loan and payment details, and suggest next steps. Other AgentCore developments include the open-source Strands harness, with planning, web retrieval and lifecycle hooks; SMART, which handles the messaging infrastructure for WhatsApp agents, helping Webmotors finish a project in six weeks instead of two months; and a Skyflow integration that tokenizes sensitive information in agent tool calls. AgentCore payments can also use AWS Billing for purchases of x402-enabled AWS Marketplace products.
Two reports put AI cybersecurity spending at the center of investor interest. Bitwise launched its actively managed AI Cyber Defense ETF, ZDAY, on Nov. 17, with a typical target of 10 to 30 holdings, at least 80% in AI-related cybersecurity companies, and a 0.75% fee. Initial top holdings include Palo Alto Networks at 13% and CrowdStrike at 11%. Gartner estimates spending on AI-amplified security could rise from $49 billion in 2026 to $204 billion by 2030. Palo Alto raised its fiscal 2027 revenue forecast to $14.52 billion–$14.57 billion and agreed to buy Applied Compute for $4.5 billion in stock; its shares rose more than 6% Thursday after gaining nearly 10% Wednesday. Other cybersecurity ETFs also advanced.
Consumers remain cautious about handing financial decisions to AI. In a survey across nine European countries, 65% said they welcome AI help with tasks such as finding information and filling out forms, but 69% were uncomfortable with AI making purchases. Trust declined from 19% for low-value purchases to 8% for high-value ones. Signicat said consumers want to retain control, with 39% seeking final approval and 32% wanting the option to override AI. The articles report different survey sample sizes—4,050 and 4,500—but the same key findings.
Debate over AI’s wider effects continued at a Federal Reserve Bank of New York conference, where Netscape co-founder and venture capitalist Marc Andreessen warned that regulation could slow innovation and limit economic benefits. He also said the benefits are not yet fully understood and called for more research. Separately, guidance on open-weight AI highlights its appeal for lower costs, privacy and control, alongside risks such as hidden malicious code, incomplete model histories and manual updates. Organizations are advised to assign model owners, review and test models regularly, and apply stronger controls to higher-risk uses.
AI is also entering security operations and customer-facing sales. Canonical lets Ubuntu Pro customers use Edge to install AI-written security patches; its Surge agents identify vulnerabilities, draft fixes for older Ubuntu versions and review them, with uncertain cases referred to Ubuntu’s security team. Kernel fixes remain outside Surge, and standard Ubuntu updates continue to receive human checks unless customers enable Edge. F5, meanwhile, issued an urgent update for 10 NGINX vulnerabilities, including two critical HTTP/2 flaws rated 9.2 that could permit remote code execution by unauthenticated attackers. It reported no evidence of exploitation and urged users, especially those with HTTP/2 enabled, to upgrade to NGINX 1.31.9.
Robotaxi operator Pony AI reported third-quarter revenue up 75% to $44.6 million, ahead of the $37.5 million analyst estimate, while its net loss widened to $105.6 million from $61.3 million, mainly due to trading-security losses. Robotaxi revenue rose 165%; its fleet numbered 2,768 vehicles at September’s end, and the company remains on track for 3,500 this year. It is testing Gen-7 robotaxis in Manchester and London and targets broader U.K. and European commercial operations in 2027. OpenAI’s first on-device AI hardware product is expected to begin shipping by year-end, with China’s Luxshare Precision as sole assembler; OpenAI has not confirmed the design, previously reported to be screenless. Jordan Belfort’s startup Tone also launched AI sales voices, offering 16 English- and Spanish-speaking voices trained on his techniques. After raising about $10 million, Tone says it has earned about $1 million and aims for $20 million–$30 million in recurring revenue within a year. Belfort says it is intended to extend sales teams, not replace them, and that companies should disclose when customers are speaking with AI.
Key Takeaways
- Amazon Bedrock AgentCore gains new agent-building, data-protection, WhatsApp messaging and AWS Billing capabilities; Webmotors completed a project in six weeks rather than two months using SMART.
- LoanPro’s multi-agent assistant summarizes borrower accounts and retrieves loan and payment details; its platform serves more than 450 lenders.
- Bitwise launched ZDAY, an AI cybersecurity ETF with a 0.75% fee; Gartner projects AI-amplified security spending will rise from $49 billion in 2026 to $204 billion by 2030.
- Palo Alto Networks raised its fiscal 2027 revenue forecast to $14.52 billion–$14.57 billion and agreed to acquire Applied Compute for $4.5 billion in stock.
- European survey respondents welcomed AI assistance (65%) but largely opposed AI making purchases (69%); trust was 19% for low-value and 8% for high-value purchases.
- Marc Andreessen warned at a New York Fed conference that regulation could slow AI innovation, while acknowledging its economic benefits remain uncertain and need further study.
- Canonical’s Ubuntu Pro Edge feature lets customers install AI-written security patches; uncertain cases go to Ubuntu’s security team, and kernel fixes remain outside Surge.
- F5 urged NGINX users to upgrade to version 1.31.9 after disclosing two critical, unauthenticated remote-code-execution flaws involving HTTP/2; it said it had found no evidence of exploitation.
- Pony AI’s third-quarter revenue rose 75% to $44.6 million, and its fleet reached 2,768 vehicles; it is testing Gen-7 robotaxis in Manchester and London.
- OpenAI’s first on-device AI hardware product is expected to ship by year-end, with Luxshare Precision as sole assembler. Tone launched 16 AI sales voices after raising about $10 million.
LoanPro builds an AI assistant to speed up loan servicing
LoanPro worked with the AWS Generative AI Innovation Center and AllCloud to build an agentic AI assistant for loan servicing agents. The assistant uses a multi-agent design with the Strands Agents SDK to summarize borrower accounts, retrieve payment and loan details, and recommend next steps. It uses Amazon Bedrock tools to connect with LoanPro systems and help agents provide more consistent service. LoanPro says its platform serves more than 450 lenders, and loans managed on it reach about 31 percent of US households.
SMART makes it easier to build AI agents for WhatsApp
SMART provides a channel layer for teams building WhatsApp AI agents with Amazon Bedrock AgentCore. It handles tasks such as authenticating and organizing incoming messages, processing media, preserving conversation memory, scheduling reminders, and managing Meta’s approved message templates for messages sent more than 24 hours after a user’s last message. The setup replaces the work of connecting more than 10 AWS services and supports local testing and deployment through Terraform. Webmotors completed a project originally planned for two months in six weeks and moved its internal team into controlled production ahead of a full launch.
Skyflow protects sensitive data in AI agent tool calls
Skyflow and Amazon Bedrock AgentCore Gateway can hide sensitive data from the tools AI agents use, then restore it for authorized systems. A Skyflow interceptor runs as an AWS Lambda function at the gateway’s request and response points, detecting information such as names, email addresses, phone numbers, government IDs, card numbers, birth dates, and locations. It replaces those values with tokens stored in Skyflow Vault. Three modes protect tool arguments and results, model inputs and outputs, or documents. The open source AWS CDK interceptor includes a Lambda function, a DynamoDB table, and Amazon CloudWatch metrics and alarms.
Amazon Bedrock AgentCore harness adds tools and lifecycle hooks
Amazon Bedrock AgentCore harness now uses Strands harness, an open source tool released under the Apache 2.0 license, and adds lifecycle hooks for custom checks and notifications. Its managed defaults include a recommended model, task-focused instructions, automatic prompt caching for supported Anthropic models on Amazon Bedrock, and proactive context management. Built-in tools support planning, web retrieval, filesystem access, context offloading, and task delegation. The changes arrive in existing AgentCore harness use without requiring a new resource or integration, while AgentCore manages secure session execution, scaling, and related services.
AI agents can buy AWS Marketplace x402 products through AWS Billing
Amazon Bedrock AgentCore payments now supports AWS Billing for AI agents buying x402-enabled products on AWS Marketplace. When a seller returns a signed quote with an HTTP 402 Payment Required response, AgentCore payments checks the agent’s session budget and returns a payment token. The agent retries the request with that token, and AWS charges the purchase to the account where Payment Manager is set up. A first purchase creates an agreement with the seller, and later purchases appear as AWS Marketplace line items. The article also introduces a sample research agent that finds a Tavily search tool, pays for one request, and provides an answer with sources.
Bitwise launches ETF focused on AI cybersecurity
Bitwise launched the actively managed Bitwise AI Cyber Defense ETF, ticker ZDAY, on Nov. 17. The fund will usually hold 10 to 30 companies and invest at least 80% in AI-related cybersecurity firms. Its largest initial holdings include Palo Alto Networks at 13% and CrowdStrike at 11%; the fund charges a 0.75% management fee. Bitwise cited Gartner estimates that spending on AI-amplified security could grow from $49 billion in 2026 to $204 billion by 2030.
Palo Alto deal fuels cybersecurity ETF gains and AI security interest
Palo Alto Networks raised its fiscal 2027 revenue forecast to between $14.52 billion and $14.57 billion, and its shares rose more than 6% on Thursday after gaining almost 10% on Wednesday. The company also agreed to buy AI startup Applied Compute in a $4.5 billion all-stock deal, expected to close in its fiscal third quarter. Cybersecurity ETFs, including BUG, HACK, and IHAK, also gained. Bitwise recently launched the Bitwise AI Cyber Defense ETF, while Gartner estimates AI-amplified security spending could rise from $49 billion in 2026 to $204 billion by 2030. ZDAY charges a 0.75% annual fee, compared with 0.58% for the broader First Trust NASDAQ Cybersecurity ETF, or CIBR.
European consumers want AI help but not control over purchases
A survey of 4,050 compliance and identity decision-makers and consumers in nine European countries found that 65% welcome AI help with tasks such as finding information, filling out forms, and exploring financial products. However, 69% are uncomfortable letting AI make purchases for them. Only 19% would trust AI with a small, low-value purchase, and just 8% would trust it with a high-value purchase. Signicat said consumers want to keep control, with 39% wanting final approval and 32% wanting the ability to override AI at any time.
Survey finds consumers wary of AI making financial purchases
A survey of 4,500 consumers and compliance and identity decision-makers across nine European countries found that 65% are open to using AI to search for information, complete forms, or explore financial products. But 69% are uncomfortable with AI making purchases for them; trust fell from 19% for low-value purchases to 8% for high-value ones. Signicat said rising identity fraud is one concern and stressed that providers need to clarify what AI can do and when consumers must approve an action. Emma Bauer said consumers welcome AI assistance but still want the final say.
Tech leaders warn AI rules could slow economic growth
At a Federal Reserve Bank of New York conference, tech executives and economists discussed AI’s possible effects on the economy, employment, and inequality. Venture capitalist and Netscape co-founder Marc Andreessen warned that regulation could slow innovation and limit AI’s economic benefits. He also said those benefits are not yet fully understood and that more research is needed. The event was held in partnership with the Economic Club of New York as part of a series on AI’s effects on the economy and society.
Ubuntu lets Pro users test AI written security patches
Canonical is allowing Ubuntu Pro customers to install security patches written by AI through a feature called Edge. Ubuntu’s regular updates will still receive human checks unless customers turn Edge on, in which case they must decide whether its patches are safe for production. Surge uses several AI agents to find relevant vulnerabilities, create fixes for older Ubuntu versions, and review the proposed patches. Uncertain cases go to Ubuntu’s security team, and approved patches run through standard tests. Kernel fixes remain outside Surge, and new vulnerabilities still follow coordinated disclosure.
Pony AI revenue tops estimates as robotaxi fleet grows
Pony AI’s third-quarter revenue rose 75% to $44.6 million, beating analysts’ estimate of $37.5 million, according to a Nov. 17 filing. Revenue from robotaxi services climbed 165%, while the company’s net loss widened to $105.6 million from $61.3 million, mainly because of losses on trading securities. Its fleet reached 2,768 vehicles at the end of September, and Pony AI remains on track to reach its target of 3,500 this year. The company has begun testing its Gen-7 robotaxis in Manchester and London, and aims for wider commercial operations in the U.K. and Europe in 2027. CEO James Peng said in August that a weaker economic outlook creates near-term challenges.
OpenAI hardware device expected to ship by year end
OpenAI’s first on-device AI hardware product is expected to begin shipping by the end of this year. China’s Luxshare Precision will be the sole assembler of the complete device and will supply core components. Several other Chinese vendors are also part of the project. OpenAI has not officially announced the product’s design, though earlier reports described it as screenless.
Open weight AI brings privacy benefits and governance risks
Organizations are adopting open-weight AI for lower costs, data privacy, and greater control, and some models may match leading provider-hosted models. But access to a model’s weights does not necessarily include its training code or training data details. Running an open-weight model on a bank’s own server can keep loan files local, while risks include hidden malicious code, gaps in a model’s history, and the need for manual updates. Organizations should set regular reviews, assign model owners, store models in internal repositories, and test them as attackers might. They should weigh each model’s benefits and risks and apply stronger controls to higher-risk uses.
F5 fixes critical NGINX flaws that could allow remote attacks
F5 issued an out-of-band security update for 10 NGINX vulnerabilities, including two critical flaws that could let unauthenticated attackers run code remotely. The critical issues, CVE-2026-123633 and CVE-2026-123007, both involve how NGINX handles HTTP/2 traffic and carry a severity score of 9.2 out of 10. The latter flaw affects versions going back to 1.9.5, released about a decade ago. F5 said it had found no evidence that attackers were exploiting the flaws, and urged organizations to upgrade to NGINX 1.31.9, especially when HTTP/2 is enabled. NGINX is widely used for websites and cloud infrastructure, with about 73 million websites reported to use it.
Jordan Belfort launches AI sales voice startup Tone
Jordan Belfort’s startup Tone launched publicly on Nov. 16 after raising about $10 million. The company has earned about $1 million so far, and Belfort hopes to reach $20 million to $30 million in recurring revenue within a year. Tone offers 16 male and female AI voices trained on his sales techniques, in English and Spanish, and is designed to help sales teams reach more people rather than replace them. Belfort said its controlled scripts may be more reliable than human salespeople, but the technology still struggles with complex conversations because it must respond within about 500 to 600 milliseconds. He expects faster chips and better AI models to narrow that limitation within six months to a year, and said companies should disclose when customers are speaking with AI.
Sources
- How LoanPro built an agentic AI loan servicing assistant with Amazon Bedrock AgentCore
- Build WhatsApp AI agents with Amazon Bedrock AgentCore
- Protect agent tool calls with Skyflow and Amazon Bedrock AgentCore Gateway
- Amazon Bedrock AgentCore harness adds Strands harness defaults, built-in tools, and lifecycle hooks
- Enable AI agents to pay for AWS Marketplace x402 products with Amazon Bedrock AgentCore payments
- Bitwise Bets on AI Security With New ETF as Spending Could Hit $204B
- Palo Alto's AI Deal Sparks Cybersecurity Rally. These ETFs Join In - First Trust NASDAQ Cybersecurity ETF…
- Survey: Consumers aren’t ready for AI agents to make purchases on their behalf
- Financial products survey highlights complexity of AI-driven purchases among consumers
- How tech leaders could shape the Fed's thinking on AI
- Ubuntu AI security patches are here, if you’re willing to test them yourself
- Pony AI’s sales beat estimates as robotaxi business expands
- OpenAI's First AI Hardware Device May Ship by Year-End, With Luxshare as Sole Assembler
- The Governance Gap in Open-Weight AI and What to Do About It
- AI finds 'internet breaker' security flaws in NGINX — open source technology responsible for running hundreds of millions of websites has decades-old issues which can be hijacked to steal data
- ‘Wolf of Wall Street’ Jordan Belfort on his AI salespeople who are better than humans: ‘What if you have a bad apple that says stuff?’
Comments
Please log in to post a comment.