1Highlight: The FBI warns AI hacking tool Artex can go from first target to admin access in under three hours.
2Philippines: Marcos signs an order creating a National AI Council and a risk-based AI governance framework.
3Also: US chip-bill push after Huang's exit, AI Rudder's phone-calling agent, and DeepSeek's cheap app APIs.
Highlight of the day
Security
<3 hrs
from first target to privileged access in some Artex attacks (FBI)
Merged from 3 sources
★ Highlight of the day
FBI warns AI hacking tool Artex can break in within three hours
The FBI issued a Flash alert A short, urgent FBI warning to organisations about an active threat, listing signs to look for and steps to take. on 13 November (US time) about Artex, an Open-source Software, or an AI model, whose code or files are published so anyone can use and change them.Penetration testing Attacking your own systems on purpose, with permission, to find weak spots before criminals do. tool that criminals and state-backed hackers are using to automate break-ins. It plugs into almost any AI model and, the FBI says, some attacks went from first target to administrator-level access in under three hours.
What we learned
Artex turns a plain-language goal into commands that AI agent An AI program that can take actions for you, like writing code, filling forms or moving items, not just answering questions. carry out, from scouting a target to stealing data. Its developer took it off GitHub in October, but the FBI says copies, forks and a paid 'Artex Pro' version keep it in circulation, with more than 1,600 Artex consoles exposed online.
The FBI says attackers mostly pair it with cheap Chinese models, led by DeepSeek's Flash family, though it also works with Claude, GPT, Grok and Kimi. It was linked to a South Korean financial attack that exposed about 67,000 people's data and to a ConnectWise backup breach reaching more than 2,250 customer accounts.
Amazon published its own study the same day: from 3 October to 5 November its Honeypot A decoy computer system set up to attract attackers so defenders can watch what they do. caught Artex agents, which Amazon links with moderate Confidence score A number, often from 0 to 1, showing how sure a model is about its answer; it does not prove the answer is right. to Chinese-linked groups, hunting for exposed AI service keys and using stolen keys to pay for more hunting. A second campaign tried Deepfake Fake video, image or audio made with AI to look or sound like a real person. to pass crypto exchanges' KYC Know your customer: identity checks that banks, crypto exchanges and similar firms must run to confirm who a customer is. identity checks; Amazon says every attempt was rejected.
The FBI's first advice is to shrink your Attack surface Every part of an organisation an attacker could try to reach from outside, such as websites, portals, apps and login pages.: list every internet-facing system, especially partner and supplier portals, old services and apps built by other departments. Then apply Least privilege Giving each person, app or key only the access it needs for its job, and nothing more., keep keys out of code and settings files, and put Rate limit A cap on how many requests or messages you can send to an AI service in a set amount of time. on business systems.
Why it matters: Attacks that once took skilled teams days can now be run by small groups at machine speed, and they look for the systems a company has forgotten about. Check your supplier portals and stray API key A secret code that lets a program use a paid online service. Anyone who steals it can run up your bill. this week rather than waiting for an incident.
FBI flash alert says Artex tool automating cyberattacks at 'machine speed' · The RecordRead ↗
Amazon Threat Intelligence observes threat actors running cybercrime campaigns with ARTEX AI agents · AWS Security BlogRead ↗
Was this useful?Tap to vote. Hold, right-click or press Shift+Enter to add a reason or a note first.
More stories this day
Philippines
4
parts of the Philippines AI Governance Framework
Philippines
Philippines sets up a national AI council to steer AI rules
Philippine President Ferdinand Marcos Jr. signed Executive order A binding instruction issued by a president or head of government without passing a new law. 131 on 12 November, adopting a national AI governance framework A set of principles, standards and rules a country or company uses to guide how AI is built and used. and creating a National Artificial Intelligence Council to put it into practice, Rappler reported on 14 November.
Key learnings
The executive secretary chairs the council and the economy and planning secretary (DEPDev) is vice chair. Members include the science, ICT, trade, finance, education, labour and budget secretaries, the higher-education and skills-training chiefs, and one private-sector representative.
The framework has four parts: ethics and principles, standards and testing, a risk-based policy toolkit, and support such as data, research and infrastructure. AI should be human-centric, transparent, explainable, safe and fair.
It is not a single AI law. Rules are meant to match the level of risk and may include sector guidelines, AI impact assessments and regulatory Sandbox A walled-off space on a computer where an AI or program can run without touching anything outside it., while existing regulators keep their current powers.
Agencies must report progress every quarter, and the council must Code review Checking code someone (or an AI) wrote before it goes live, to catch bugs and risky changes. the framework at least every two years, sooner if technology or markets change sharply. The order takes effect once published.
Why it matters: The Philippines joins neighbours such as Singapore and Malaysia in choosing guidelines and one coordinating body over a broad AI law. Companies working across ASEAN should expect risk checks and impact assessments rather than outright bans.
Marcos creates National Artificial Intelligence Council · RapplerRead ↗
Was this useful?Tap to vote. Hold, right-click or press Shift+Enter to add a reason or a note first.
Chips & policy
3
US chip-control bills lawmakers now expect to move
Merged from 4 sources
Chips & policy
After Huang's exit, US lawmakers push chip bills as Asia reassures
A day after Jensen Huang quit as Nvidia's chief executive, US lawmakers said tighter chip-Workspace export A download of everything in a Notion workspace as files, often used as a backup. Today it can't be loaded straight back into Notion. bills now look likely to pass, and governments across Asia moved to calm nerves. TechCrunch rounded up the Asian reactions on 14 November (US time).
Key learnings
The AI Daily Brief podcast (14 November episode) quoted House Foreign Affairs chair Brian Mast saying the scandal had '100%' boosted the chances of three Chip-control bill A proposed law limiting which countries or companies can buy, receive or rent advanced computer chips.. Citing Politico, it said OpenAI has privately told congressional offices it supports all three, and citing Bloomberg, that Huang agreed to testify at a closed-door House hearing this month.
One of the bills, the Remote access Controlling a computer from somewhere else over the internet, which scammers try to get to steal data. Security Act (RASA), would treat renting US chips over the Cloud computing Renting computers in someone else's data centre over the internet instead of buying your own. like exporting them. On the ChinaTalk podcast, host Jordan Schneider said this aims to close the loophole Chinese firms use by setting up cutouts in Malaysia. SemiAnalysis' Dylan Patel argued against RASA, saying US-controlled rentals that can be switched off are safer than pushing China to smuggle more chips.
In Asia: South Korea says Nvidia confirmed its plan to supply 260,000 advanced GPU A powerful computer chip, originally for graphics, that is now the main hardware used to train and run AI., Taipei says Nvidia's new Taiwan headquarters is on track, and Japan says it will monitor closely. Singapore's home ministry says its Megaspeed investigation, which looked at whether servers shipped to Malaysia were sent on without proper declaration, is still open and covers six other people.
Inside Nvidia, CFO Colette Kress is Interim CEO A temporary chief executive who runs a company until a permanent one is chosen.. Lowyat.net, citing reports, says the board has approached Arm CEO Rene Haas; the board calls talk of candidates premature. An internal post asking for Huang's return has drawn more than 1,600 comments.
Why it matters: If RASA-style rules pass, renting AI computing power to Chinese customers from Malaysian or other Southeast Asian Data centre A large building full of computers that store data and run online services and AI, and that uses a lot of power and water. could need the same US approval as shipping the chips. Operators and tenants here should check who their end users are now.
NVIDIA Reportedly Eyes ARM CEO As Potential Successor To Jensen Huang · Lowyat.netRead ↗
Was this useful?Tap to vote. Hold, right-click or press Shift+Enter to add a reason or a note first.
AI agents
100
free credits for new CALL-E accounts
Merged from 3 sources
AI agents
Singapore's AI Rudder lets AI agents make phone calls after approval
Singapore-based AI Rudder launched CALL-E on 13 November, a Voice agent An AI system that holds spoken conversations, for example making or answering phone calls. that lets AI assistants such as ChatGPT, Muse and Claude Code phone a restaurant, clinic or customer for you. Each call goes ahead only after you approve its plan.
Key learnings
How it works: your AI agent An AI program that can take actions for you, like writing code, filling forms or moving items, not just answering questions. hands CALL-E the goal and the phone number. CALL-E shows you the call plan, dials only after you approve, then reports what was agreed, how confident it is, the exact lines from the call that back this up, and a transcript.
Calls introduce themselves as an AI voice agent by default. You can change the greeting, but a pre-call check Block (Notion) A single piece of content on a Notion page, such as a line of text, a heading, an image, or a database row. pretending to be emergency services or the police.
AI Rudder says CALL-E works with ChatGPT, Muse, OpenClaw, Hermes Agent, Claude Code, Codex OpenAI's coding agent that can write, change and test software from written instructions. and Cursor. TechNode Global notes this is AI Rudder's own claim, not a partnership with those companies, and its interruption-handling Benchmark A standard test used to compare AI models, like an exam everyone takes so scores can be compared. result has not been checked independently.
New accounts get 100 free credits; after that you pay by call length plus phone charges. AI Rudder already runs enterprise voice agents in Malaysia, Indonesia, the Philippines and China.
Why it matters: Phone calls are one of the last errands AI agents could not do, and the approve-first, prove-after design is a sensible guard. Before using it with customers, check local consent and recording rules: your approval does not replace the other person's consent.
Was this useful?Tap to vote. Hold, right-click or press Shift+Enter to add a reason or a note first.
Developer tools
RM0.24
about the price of one hour of speech recognition (half-price offer)
Merged from 2 sources
Developer tools
DeepSeek opens its app's search, file and voice tools to developers
DeepSeek on the evening of 13 November (China time) opened the features behind its own app to developers: web search, file upload, link reading, text recognition and, for the first time, Speech recognition Technology that turns spoken audio into written text. and Text-to-speech Technology that reads written text aloud in a computer-generated voice.. It says developers could rebuild the whole DeepSeek app with them.
Key learnings
Each tool has its own interface and price, so you can use just one. With web search switched on, the model now decides by itself when to search, what to read and how to answer, instead of developers wiring those steps together.
Prices from DeepSeek's list, reported by QbitAI: speech recognition 0.39 yuan an hour (about RM0.24) during a half-price offer, text-to-speech 1.5 yuan per 10,000 characters, OCR Optical character recognition: software that reads words in pictures or scans and turns them into text you can edit. 1.5 yuan per 1,000 pages, and search or page fetching 5 yuan (about RM3) per 1,000 successful calls.
QbitAI's own comparison puts DeepSeek's transcription at roughly a third to a sixteenth of the price of OpenAI's and Google's speech services.
The features run on its V4.1-Flash model across its three API Application programming interface: a door that lets one piece of software ask another for data or actions. formats, including ones compatible with OpenAI and Anthropic tools. DeepSeek also open-sourced a simple chat front end that can import your DeepSeek chat history.
Why it matters: Cheap building Block (Notion) A single piece of content on a Notion page, such as a line of text, a heading, an image, or a database row. make it easier for small teams to build voice and document assistants. Note that the FBI alert in today's top story names DeepSeek Flash models as attackers' favourite with Artex, so check how any provider handles your data and misuse before you build on it.