Friday, 20 November 2026 Follow @AIDailyFive

The AI Daily Five · No. 29 · Wednesday 18 November

How Fragile Is America’s Tech Power?

From an NSA network failure and new US limits on remote access to AI computing to Nvidia’s record quarter without its founder, OpenAI’s more capable models and Bitcoin under pressure, today’s edition asks how resilient America’s technological power really is.

A gold Bitcoin coin leaning beside a red arrow falling across a chart of red trading candles.
Illustration: AI Story Front design team

1

NSA Network Failure Exposes US Digital Fragility

America’s intelligence agency has joined the growing list of institutions disrupted by technical failures. Officials say there is no evidence connecting the incidents to cyberattacks, but the consequences extend well beyond inconvenience.

The National Security Agency’s classified internal communications network went offline on Tuesday, 17 November, disrupting work at one of the United States’ most sensitive intelligence organisations.

The system, known as NSANet, allows agency employees to exchange classified information. According to CNN’s original reporting, the disruption prompted staggered arrival times for personnel while officials assessed the problem.

The NSA subsequently attributed the incident to routine maintenance, describing a temporary degradation of some internal systems. It said critical national-security capabilities remained operational.

The outage was the latest in a series of technical failures affecting essential American services.

A Difficult Week for US Infrastructure

A failure involving the Federal Aviation Administration grounded flights across the country on Friday, disrupting thousands of passengers.

Officials traced that incident to a router problem in Arkansas.

Microsoft services also suffered interruptions, affecting applications including Outlook and Teams. Separately, technical problems at JPMorgan Chase created difficulties for financial transactions, while United Airlines experienced an internal systems failure that delayed flights.

The incidents occurred within days of one another, prompting speculation about whether they were connected.

Speaking during a visit to the Philippines, Secretary of State Marco Rubio rejected suggestions that the United States was experiencing a coordinated attack.

According to CBS News, Rubio said investigators had found no evidence that the failures were related or caused by hackers, criminals or foreign governments.

His comments did not establish that every technical fault had been fully resolved. Rather, they distinguished a series of apparently separate operational problems from evidence of hostile activity.

What Happened at the NSA?

The agency’s statement reported by CBS News attributed the interruption to scheduled maintenance.

Officials said the NSA had multiple layers of redundancy designed to maintain critical operations during system failures.

Current and former US officials also told CNN that the disruption would not prevent urgent intelligence from reaching President Donald Trump during his trip to Asia.

Nevertheless, the outage affected an internal network central to the agency’s everyday work.

Even systems protected by extensive security measures remain dependent on software maintenance, hardware reliability and the people responsible for keeping them running.

2

Trump Meets Xi as US Tightens AI Chip Controls

The US president arrives in China for talks with Xi Jinping and Vladimir Putin as Washington introduces restrictions on access to advanced computing infrastructure. The dispute over AI technology now extends beyond physical chips.

US President Donald Trump is preparing to meet Chinese President Xi Jinping and Russian President Vladimir Putin during the Asia-Pacific Economic Cooperation summit in Shenzhen, China.

The gathering comes amid intensifying disagreements over trade restrictions, technology access and the international security order.

According to Reuters reporting on 18 November, Xi criticised protectionism in remarks to the summit, while Putin challenged Western sanctions.

Trump’s planned meeting with the two leaders is expected on Thursday, 19 November. Among the issues surrounding the talks are Russia’s war in Ukraine, the war involving Iran and the restrictions governing advanced technology.

An important development emerged from Washington just two days before the summit.

The US Closes Another Route to AI Computing Power

On 16 November, the US Department of Commerce announced an interim final rule addressing remote access to advanced computing technology.

Existing American export controls restrict the shipment of certain advanced chips and computing systems to designated countries and entities.

But physical ownership is not the only way to use computing power.

A company unable to purchase restricted chips directly may be able to access similar capabilities through a cloud provider operating in another country.

The Commerce Department says the new rule is designed to prevent sensitive American technology from being accessed through remote computing arrangements that circumvent existing restrictions.

This broadens the practical reach of export controls from the physical movement of hardware towards the provision of computing services.

Why Cloud Computing Complicates Enforcement

Cloud infrastructure rarely operates within the economic boundaries of a single country.

A data centre may be located in one jurisdiction, owned by a company headquartered in another and accessed by customers elsewhere.

Restricting the sale of advanced chips is therefore different from controlling who can use the computational power those chips provide.

The new US measures seek to address that distinction.

They also create additional compliance questions for cloud providers serving international customers, although the precise obligations depend on the rule’s legal provisions and the parties involved.

China has repeatedly criticised US restrictions on advanced technology, arguing that they interfere with international trade and the development of its domestic technology industry.

Washington maintains that its controls are intended to protect national security.

A Summit With Competing Priorities

The APEC gathering offers an opportunity for the leaders to address several international disputes.

Trade and technology are significant parts of that wider agenda, but the meetings are also expected to address geopolitical and military tensions.

No agreement on AI export controls has been announced.

The timing of Washington’s latest measures nevertheless illustrates how closely advanced computing has become connected to international diplomacy.

3

Nvidia Doubles Revenue but Rejects Huang’s Return

The AI chipmaker has reported another record-breaking quarter, with revenue more than doubling from a year earlier. Yet its financial success contrasts sharply with a leadership controversy that has reached Washington.

Nvidia reported quarterly revenue of $115.8 billion on Tuesday, 17 November, an increase of 103% from the same period last year.

The results exceeded the company’s previous quarterly performance, with data-centre sales reaching $107.6 billion, up 110% year on year.

In its official financial results, Nvidia forecast approximately $130 billion in revenue for the following quarter, plus or minus 3%.

These figures demonstrate the extraordinary demand for the infrastructure supporting artificial intelligence.

But the company’s earnings announcement was accompanied by a different development: Nvidia’s board confirmed that former chief executive Jensen Huang would not be reinstated.

Nvidia Rules Out Huang’s Return

Huang resigned following revelations about his personal conduct and scrutiny surrounding his relationship with a Chinese businesswoman whose activities have attracted national-security concerns.

The circumstances have generated investigations and political pressure, although allegations involving espionage and possible technology diversion have not been established as findings against Huang.

According to Associated Press reporting on 18 November, Nvidia chairman Harvey Jones confirmed that the board had unanimously accepted Huang’s resignation and did not intend to reinstate him.

Huang had offered to return, while President Donald Trump publicly supported his reinstatement.

The board’s decision leaves Colette Kress, Nvidia’s chief financial officer, leading the company on an interim basis.

Nvidia has also announced the creation of a board committee responsible for national-security and export-control oversight.

The company is cooperating with congressional inquiries, with Huang having agreed to transcribed interviews with House lawmakers.

AI Demand Continues Despite the Controversy

The leadership crisis has so far done little to diminish the company’s reported financial performance.

Nvidia’s third-quarter revenue increased 20% from the preceding quarter, while reported net income reached $65.3 billion.

The company also announced that its Vera Rubin NVL72 computing platform had begun shipping.

Nvidia’s fourth-quarter forecast assumes no data-centre computing revenue from China, underscoring the commercial consequences of US technology restrictions.

Interim chief executive Kress emphasised the company’s technology roadmap and operational continuity.

What Comes Next?

Nvidia’s immediate challenge is to preserve its technical and commercial momentum while establishing permanent leadership.

The company also faces increased scrutiny over its compliance arrangements and relationships involving China.

Its financial results suggest strong demand for advanced computing infrastructure. They do not resolve the governance concerns that have emerged following Huang’s resignation.

4

OpenAI Launches GPT-6.2 With Higher Safety Risks

OpenAI’s latest models promise stronger professional performance and lower costs. Its accompanying safety assessment also identifies advanced cybersecurity and AI self-improvement capabilities, complicating the argument for wider deployment.

OpenAI has released GPT-6.2 Astra and GPT-6.2 Sol, two artificial intelligence models designed for different levels of professional and everyday work.

Announced on 17 November, Astra targets complex tasks involving software development, scientific research and extended professional workflows. Sol offers a less expensive alternative for common tasks requiring text processing, images and tools.

Both models are available through supported ChatGPT services, Codex and the OpenAI API.

The release comes with claims of improved performance and efficiency, but also significant disclosures about the models’ potential risks.

What Has Improved?

OpenAI says GPT-6.2 Astra achieves stronger results than its predecessor on several professional benchmarks.

On Agents’ Last Exam, a benchmark measuring multi-step professional tasks, Astra scored 5.2 points higher than GPT-6 Astra while completing the evaluated work at less than one-third of the cost.

The company also reports improvements in coding, computer use and the production of professional documents.

Astra’s standard API pricing is $8 per million input tokens and $40 per million output tokens.

Sol costs $2 and $10 respectively.

OpenAI says Astra’s lower cost per task results partly from using fewer output tokens to complete work, rather than simply lowering the price charged for each token.

These are company-reported benchmark results. They indicate improvements on selected evaluations but do not establish comparable performance across every professional setting.

The Safety Assessment Deserves Attention

The most consequential information may be found in OpenAI’s accompanying GPT-6.2 system card.

Under the company’s Preparedness Framework, both Astra and Sol meet its Critical capability threshold for cybersecurity and its High threshold for biological and chemical capabilities.

Astra also meets the High threshold for AI Self-Improvement, making it the first externally released OpenAI model to reach that classification.

This does not mean the model has independently improved itself or escaped human supervision.

The classification concerns capabilities measured during safety evaluations, including abilities relevant to accelerating AI research and development.

OpenAI says it has introduced additional safeguards, including tighter restrictions on higher-risk cybersecurity assistance and more stringent access requirements for certain uses.

What Does This Mean for Users?

For organisations using AI to write software or perform research, improved capabilities could reduce the time and expense associated with complex tasks.

However, increasingly capable models also require more careful assessment of where and how they are deployed.

OpenAI’s announcement illustrates both aspects of the development: improved professional performance and a higher level of risk identified by the company’s own evaluations.

5

Bitcoin Falls as ETF Inflows Return, Briefly

Bitcoin has slipped towards $65,000 after another difficult week for digital assets. Yet a modest return of money to US exchange-traded funds complicates the picture of an uninterrupted institutional retreat.

Bitcoin remained under pressure on Wednesday, 18 November, following a sharp decline that pushed the cryptocurrency towards $65,000.

The fall reflects a broader deterioration in investor sentiment, with digital assets facing selling pressure alongside uncertainty in technology markets.

However, fresh data from US spot Bitcoin exchange-traded funds presents a more complicated picture than the falling price alone suggests.

After seven consecutive trading sessions of net withdrawals, the funds recorded a modest net inflow on Monday, 16 November.

Did Institutional Investors Stop Selling?

According to the TFTC Bitcoin ETF flow tracker, US spot Bitcoin ETFs attracted approximately $8 million in net inflows on 16 November.

The reversal followed a period of substantial withdrawals.

The previous trading session, 13 November, recorded approximately $185.5 million in net outflows. On 12 November, withdrawals totalled about $284 million.

The monthly figures remain considerably more negative.

Through the latest reported session, November’s cumulative net withdrawals stood at approximately $2.34 billion.

The return to positive flows on one day therefore represents a limited improvement rather than clear evidence of a sustained recovery.

BlackRock’s IBIT fund recorded approximately $76.5 million in inflows on 16 November, partially offset by withdrawals from several competing products.

The figures illustrate how different institutional investment vehicles can experience opposing movements even when the overall market appears to be moving in one direction.

Why Is Bitcoin Under Pressure?

Bitcoin’s decline has coincided with broader concerns about technology valuations and risk appetite.

The cryptocurrency has also fallen below its 200-week moving average, a long-term market indicator followed by some traders.

Such indicators can influence trading behaviour, but they do not independently establish where an asset’s price will move next.

The recent sell-off has affected Ethereum and other digital assets, reflecting the close relationship between cryptocurrency prices and investors’ willingness to hold volatile investments.

Bitcoin’s movements also demonstrate that the expansion of regulated investment products has not eliminated the market’s sensitivity to changes in sentiment.

What Do the ETF Figures Tell Us?

Exchange-traded funds have become an important route through which institutional and retail investors gain exposure to Bitcoin.

Net fund flows provide information about changes in holdings through those products. They do not, however, represent every transaction in the underlying Bitcoin market.

An individual day of inflows can coexist with falling prices if selling pressure elsewhere is stronger.

Similarly, substantial withdrawals over several weeks do not establish that every institutional investor is abandoning the asset.

The distinction matters when interpreting a market in which different types of participants respond to different financial pressures.

Compiled with AI assistance. Written and checked by a human editor before publication.