Kong is expanding its AI platform with new cost controls, observability and governance tools. Its Token Vault also helps secure AI agent access to enterprise resources.

Read more: buff.ly/CSFCVap

#AgenticAI #AIGovernance #Observability

With AI fever engulfing almost everyone, the GPUs on developers' computers are doing a lot more work these days, which is why Microsoft technical executive Scott Hanselman built GPU Taskman, a Python tool he likens to Linux's htop that shows what is actually occupying your GPU from the terminal.

It separates model weights from KV caches and compute buffers, pulls context window sizes and throughput metrics out of local servers like Ollama and LM Studio, and runs on Windows, WSL, Linux, Apple, and even recognizes containers on Linux.

Have a look at what it tracks and why local GPU monitoring has been so underserved: buff.ly/GIybyUy

#AI #GPU #DeveloperTools #LocalAI #Observability #OpenSource

A knockoff handbag so close to the real thing that the original was purchased online and the $175 replica mailed back for a refund is exactly the scheme Happy Returns now uses cameras, computer vision and its Return Vision software to catch, comparing flagged merchandise against the merchant's catalog photos for the wrong logo, the too-wide strap, the fabric that doesn't match.

Fraud is the dark side of convenience, as its CEO puts it, and with roughly 9% of all U.S. returns fraudulent and more than $75 billion in merchandise at stake, the schemes are growing in sophistication as fast as the tools built to spot them.

Read more about how retailers are separating legitimate returns from deliberate deception: buff.ly/0yWFWwP

#AI #Retail #ComputerVision #FraudPrevention #Ecommerce

Congressional Democrats are preparing a legislative campaign targeting AI and the rapid expansion of energy-hungry data centers when they take control in January, with measures to shield households from utility bills tied to data facilities, mandate independent safety audits for frontier models, and strip legal shields protecting AI developers from product liability.

Semiconductor stocks fell sharply on the news, with the iShares Semiconductor ETF dropping nearly 9% and AMD sinking more than 14%, though any statutory push still faces a president who has resisted mandatory guardrails and a veto that looms over standalone regulatory bills.

The full picture on what Democrats can actually pass, and what the market is already pricing in: buff.ly/CkykpxU

#AI #AIPolicy #DataCenters #TechPolicy #Election2026

Thinking Machines Lab, founded by former OpenAI CTO Mira Murati, has released Inkling 2 Small, a second-generation model for autonomous coding, visual reasoning and knowledge work, and is reportedly nearing a $5 billion funding round at a $40 billion pre-money valuation.

Artificial Analysis gave it a score of 31 on its Intelligence Index, an 18-point improvement over its predecessor with the largest gains in agentic tasks, though its AutomationBench-AA score rose from 5% to 46% while generating roughly four times as many output tokens per task, pushing average cost to $1.18 against $0.20 for the earlier Inkling Small.

See how the benchmarks compare and what the reported round would mean: buff.ly/1UQwSMd

#AI #ThinkingMachines #MiraMurati #LLM #AIModels

Gartner is forecasting that nearly a third of workers terminated and replaced by AI will be rehired by 2029, often at higher salaries, because companies are rushing into AI-driven layoffs without understanding the operational mechanics required to sustain their businesses.

Just 19% of the nearly 300 chief HR officers surveyed reported having workforce planning frameworks capable of handling AI's impact, which is why Gartner's advice is blunt: stop planning for the workforce you have and start planning for the one your strategy requires.

Learn what the five-stage framework changes about AI-driven layoffs: buff.ly/Vo1F4ak

#AI #FutureOfWork #Gartner #WorkforcePlanning #HR

The Center for a New American Security warns that the next big AI risk is a class of sovereign agents able to acquire compute, replicate across networks and resist shutdown, creating a "proliferation with no kill switch" scenario that would mean a catastrophic loss of control.

Its report lays out seven capabilities that compound from minimal to expanded sovereignty, and notes the timing is not theoretical, since Anthropic amended its own filing on November 5 to say it is on the cusp of recursive self-improvement.

See what CNAS recommends for chokepoints, liability and international caps on open-weight models: buff.ly/keLfM89

#Cybersecurity #AIAgents #AISafety #AIGovernance

MIT researchers have figured out how to freshen up a model without expensive retraining cycles, using a technique they call weight-space programming that lets a general language model read another network's raw weights like source code.

In one test, GPT-6 Astra diagnosed a model stuck in a repetitive reasoning loop and fixed it by editing a handful of internal weights, jumping its math score from 8% to 48% with no training loop at all.

The full breakdown of how it works, and where it still falls short, is here: buff.ly/v8bTTCc

#AgenticAI #MIT #MachineLearning #LLM #AIResearch #ModelOptimization

Dropbox has extended the scope of its cloud storage service by adding DBX Agent Cloud, a platform that gives organizations AI agents capable of finding relevant content, proposing a plan and completing approved tasks across files already stored there.

As chief product officer Mike Torres puts it, you can bring the file to the model or vice versa, and Dropbox plans to acquire Tigris Data so it can store what both humans and agents create at scale while keeping an audit trail for every action an agent performs.

The full story on how the platform handles context, connectors and undo: buff.ly/yeUcvaK

#AI #Dropbox #AIAgents #CloudStorage #EnterpriseAI

Millions of people use AI models every day with no way to verify which model is actually generating their answers, and Anthropic's answer is FlashProve, an open-source research prototype that uses zero-knowledge proofs to let customers or auditors confirm an output matches a committed set of model weights without ever seeing them.

In tests with Meta's 400-billion-parameter Llama 4 Maverick it processed about 110 tokens per second on a single H100, though generating proofs cost 70 to 95 times the GPU time of conventional inference, contexts are capped at 1,024 tokens, and proofs arrive after delivery, so it detects tampering rather than preventing it.

Read more here: buff.ly/2UbzURX

#AgenticAI #Anthropic #Cryptography #AIGovernance #AISecurity

Elon Musk contributed more than $445 million to Republican candidates and allied political organizations ahead of this election, the largest donor of the cycle, and the outcome was a Democratic takeover of both chambers of Congress.

Money bought an enormous campaign; it did not deliver the result, and the same limit on the industry's power showed up where data centers appeared directly on ballots, with restrictions winning 78.6% of the vote in Sunbury, Ohio and 82.4% in Granville.

Read the full article on what these votes mean for AI infrastructure and the industry's relationship with the public: buff.ly/Cu3CJIO

#AgenticAI #AIPolicy #DataCenters #TechPolicy #Midterms2026

Google faces backlash over plans to tie Search access to AI features in schools. Districts opting out of AI could lose access to Google Search entirely.

Read more: buff.ly/4IFL107

#Google #AI #EdTech

Looking for your next AI role? This week's roundup features 10 openings at Airbnb, Waymo, HackerOne and more, with salaries up to $302K.

Explore the jobs: buff.ly/Qkf4TjF

#AI #MachineLearning #Hiring

OpenAI plans to cut data center power use during peak demand through its new Grid Protectors Program. The initiative aims to ease grid strain amid growing community opposition.

Read more: buff.ly/aFXLgTz

#OpenAI #AI #DataCenters

Salesforce has agreed to acquire Cresta, reportedly for $2.7 billion, to give Agentforce voice AI capable of holding phone conversations, answering customer questions and completing service requests without human intervention.

The harder question the deal raises is how you measure whether automated customer service actually saves money, since a call an AI handles without transferring can still be expensive if the customer spends several minutes getting nowhere.

Read the full article on what Cresta brings and where integration gets complicated: buff.ly/fLOCE8v

#Salesforce #Agentforce #AIAgents #CustomerService #VoiceAI

Elon Musk's name isn't on today's ballot, but he has put roughly $446 million behind the effort to elect Republicans, and that seems like a reasonable place to begin any discussion of technology in this election.

We can catalog the 41 data center measures in 10 states and the license plate reader questions voters will decide today, but an examination that stops there leaves out the extraordinary effort by technology's wealthiest owners to help choose the government that will oversee their businesses.

Read the full piece on who gets to call the shots, then get out and vote today: buff.ly/Ln71oM4

#AI #TechPolicy #Election2026 #DataCenters #AIGovernance

The appeal of open-weight AI is easy to understand, but an organization comparing a ChatGPT or Claude subscription with the cost of running a downloadable model is comparing different purchases, since the subscription includes an application and the services around it while the model provides a component from which an application can be built.

A new Techstrong Special Report looks at cost per successful outcome rather than the token bill, and its illustrative scenario shows an extra percentage point of documents needing four minutes of human review costing roughly $2,933 a month.

Read the full report on connecting model selection to work you actually need completed: buff.ly/yn7mExU

#AgenticAI #OpenWeights #EnterpriseAI #Techstrong #AIEconomics

Goldman Sachs is cutting off bankers and traders from AI coding tools over security and compliance risks. The move signals a shift toward tighter AI governance on Wall Street.

Read more: buff.ly/F6wytlr

#AI #GoldmanSachs #AIGovernance

Corporate leaders worldwide are rapidly adopting AI to accelerate daily decision-making, yet only 25% are using the technology to fundamentally reinvent their organizations, according to a Deloitte survey of 3,500 senior executives that paints a picture of an enterprise landscape caught between rapid short-term efficiency gains and the harder structural work required for lasting transformation.

Financial oversight remains incomplete too, with 71% monitoring aggregate AI spending but just 44% calculating granular per-unit inference costs, leaving finance departments with little insight into what's driving their escalating bills.

Read the full findings on where enterprise AI transformation actually lags: buff.ly/tFxkWA6

#AI #EnterpriseAI #Deloitte #DigitalTransformation

Cognizant is launching an OpenAI Frontier Capability Group of more than 25,000 professionals to help enterprises deploy ChatGPT, Codex and other AI tools, focused on software development, legacy IT modernization and AI-powered customer operations.

The move follows its commitment of 18,000 employees to Google's Gemini, which positions the services firm to serve customers whichever AI platform they pick.

Read the full article on what the group will tackle and where the training stands: buff.ly/yBHoobt

#AI #Cognizant #OpenAI #EnterpriseAI #Codex