人工智能並非最大的網絡安全問題。最大的問題是人(1/2)

Recently CNN reported the following:

AI isn’t the biggest cybersecurity problem. People are (1/2)

BUSINESS TECH

CNN - By Lisa Eadicicco

AUG 9, 2026

Cases of AI escaping the lab, infiltrating other companies and trying to deceive people have all made headlines in recent weeks. And in one case, AI models even worked together to break free from their test environments.

Does this mean the machines are taking over? Not quite.

AI isn’t the mastermind behind today’s most widespread cyber threats; it’s people who can use AI nefariously – and for nefarious purposes. AI has given bad actors massive power, allowing them to create malicious software, research targets, create convincing schemes and automate attacks at an unprecedented pace.

One in four data breaches were driven by AI from February 2025 to March 2026, according to an IBM report. And Americans lost more than $893 million to AI-related scams last year, the FBI says.

But experts say real-world threat actors – that is, people – are still the ones pulling the strings. AI agents have only perpetuated existing attack methods, like phishing and malware scams, rather than creating wholly new ones.

“It’s the humans that we need to watch out for,” said Oren Etzioni, professor emeritus at the University of Washington and former CEO of the Allen Institute for Artificial Intelligence. “AI is just the tool.”

AI going rogue

Some recent incidents have shown what AI is capable of in the real world, not just in theory, igniting fears about whether the technology is advancing too quickly.

In July, OpenAI test models escaped their constraints and hacked into other companies’ systems during an internal evaluation.

Days later, Anthropic said it discovered its AI models had breached three companies during testing.

In a separate test, Anthropic’s most advanced model used fake identities to try to deceive real people, researchers at Britain’s AI Security Institute said Tuesday.

One of Meta’s AI models also broke into an external company, the social media giant said Wednesday.

Those breaches also show how unpredictable AI can be when interpreting instructions. OpenAI’s models, for example, were trying to pass a cybersecurity test when they broke out of their test environment and breached another company, even though they weren’t told to do so.

Patrick Fussell, global head of adversary simulation at IBM, compared AI to a genie.

“You want to ask it a wish, but you have to be very, very specific about the details of your wish,” he told CNN. “Or it could sort of go awry.”

But these instances also happened under very specific circumstances.

OpenAI turned off restrictions that would have prevented its model from “pursuing high risk cyber activity.” Anthropic ran its tests without the standard safety safeguards in models that are generally available to the public. The AI Security Institute also turned off certain tools that could have “reduced the scope” of what the models were capable of.

Researchers typically do this to fully evaluate a model’s capabilities.

“I couldn’t go into ChatGPT or Claude or something like that, and it accidentally breaks into the FBI. That’s not going to happen,” said Adam Meyers, head of counter adversary operations at cybersecurity firm CrowdStrike. “So what we’re seeing is these are done in specific test conditions where they’re monitoring to see, ‘Does this thing do something that it’s not expected to do?”

(to be continued)

Translation

人工智能並非最大的網絡安全問題。最大的問題是人(1/2)

近幾週來，人工智能逃逸實驗室、滲透其他公司並試圖欺騙人類的案例頻頻登上新聞頭條。在其中一件案例中，人工智能模型們甚至協同合作，成功突破了測試環境。

這是否意味著機器正在接管一切？並非如此。

人工智能並非當今最普遍的網絡威脅背後的主謀；是人才會惡意使用人工智能，並用於惡意目的。人工智能给不法分子提供了巨大的能力，使他們能夠以前所未有的速度創建惡意軟件、研究攻擊目標、設計逼真的騙局並自動化攻擊。

IBM 的一份報告顯示，從2025年2月到2026年3月，四分之一的資料外洩事件是由人工智能驅動的。美國聯邦調查局（FBI）表示，去年美國人因人工智能相關的詐騙損失超過8.93億美元。

但專家表示，現實世界中的威脅行為者 - 也就是人 - 仍然是幕後操縱者。人工智能代理只是延續了現有的攻擊手段，例如網絡釣魚和惡意軟件詐騙，而不是創造了全新的攻擊方式。

華盛頓大學榮譽教授、亦是 Allen人工智能研究所 的前首席執行官Oren Etzioni 說:“我們真正需要警惕的是人”，“人工智能只是工具。”

人工智能失控

最近發生的一些事件表明，人工智能在現實世界中的能力遠超理論，引發了人們對這項技術發展是否過快的擔憂。

今年7月，OpenAI 的測試模型在內部評估期間突破了限制，入侵了其他公司的系統。

幾天後，Anthropic 公司表示，他們發現其人工智能模型在測試期間入侵了三家公司。

英國人工智能安全研究所的研究人員週二表示，在另一項測試中，Anthropic 公司最先進的模型使用虛假身份試圖欺騙真實人物。

社群媒體巨頭 Meta 公司週三表示，該公司的一款人工智能模式也入侵了一家外部公司。

這些這些違規行為也表明，人工智能在解讀指令時可能具有不可預測性。例如，OpenAI 的模型在試圖通過網路安全測試時，突破了環境測試而入侵了另一家公司，儘管它並未被告知要去這樣做。

IBM 的全球對手模擬主管Patrick Fussell將人工智能比喻為阿拉丁神燈裡的精靈。

他告訴 CNN:“你想向它許願，但你必須非常非常具體地說明你的願望” ;“事情可能或者會有點出錯。”

但這些事件也發生在非常特殊的情況下。

OpenAI 關閉了原本可以阻止其模型「進行高風險網路活動」的限制。 Anthropic 公司在進行測試時，並未採用通常對外開放的標準模型安全防護措施。人工智能安全研究所也關閉了一些可能「限制」模型功能範圍的工具。

研究人員通常會這樣做，以便全面評估模型的能力。

網絡安全公司 CrowdStrike 的反恐行動主管 Adam Meyers 說: 「我不可能在進入 ChatGPT 或Claude 之類的系統後，會意外地入侵了FBI。這種情況絕對不會發生」; 「所以我們看到的是，這些測試都是在特定測試情况監控下發生的，目的是看看‘這個系統是否會做出一些意料之外的事情？’」。

（待續）

Note:

1.  In cybersecurity, 'adversary simulation' (對手模擬) refers to mimicking the actions and tactics of potential attackers so as to test and improve an organization's defenses.

President Zelenskyy: "North Korea Decides to Send 30,000 to 50,000 Troops to Russia"

Recently NHK News on-line reported the following:

ゼレンスキー大統領“北朝鮮3万人から5万人ロシアに派兵決定”

2026年8月9日 11:17

ウクライナ情勢

ウクライナのゼレンスキー大統領は、北朝鮮が3万人から5万人の兵士をロシアに派兵することを決めたという見方を示した上で、今後、ロシアからあらゆる種類の装備を受け取ることになるとして、ウクライナへの支援を呼びかけました。

ウクライナのゼレンスキー大統領は8日、SNSへの投稿で、北朝鮮軍の動向について「ロシア領内に現れた兵士は当初数百人だったが、その後数千人規模となり、現在では3万人から5万人が派兵されることが決定している」という見方を示しました。

また、ウクライナでは北朝鮮製のミサイルが見つかっているとした上で、北朝鮮が現代戦の経験を蓄積し、ロシアから軍事技術やあらゆる種類の装備を受け取ることになるのは明らかだとしました。

そして、こうした状況はアジアの国々にとって脅威となる可能性があると指摘した上で、「韓国はウクライナと緊密に協力するべきだ」として、防空システムやドローンなどの分野での支援を呼びかけました。

北朝鮮によるロシアへの追加の派兵をめぐって、ゼレンスキー大統領はことし6月以降、ウクライナに隣接する西部ボロネジ州で受け入れ準備を進めていると指摘していました。

Translation

President Zelenskyy: "North Korea Decides to Send 30,000 to 50,000 Troops to Russia"

August 9, 2026, 11:17 AM

Ukraine Situation

Ukrainian President Zelenskyy stated that North Korea had decided to send 30,000 to 50,000 soldiers to Russia and would receive all kinds of equipment from Russia,  he called for support for Ukraine.

On August 8, President Zelenskyy posted on social media that regarding the movements of the North Korean military, "Initially, there are several hundred soldiers appearing in Russian territory, but that number has since increased to several thousand, and now it has been decided that 30,000 to 50,000 will be sent."

He also stated that North Korean-made missiles had been found in Ukraine, and that it was clear that North Korea would accumulate experience in modern warfare and receive military technology and all kinds of equipment from Russia.

He then pointed out that this situation could be a threat to Asian countries, saying that "South Korea should cooperate closely with Ukraine" and called for support in areas such as air defense systems and drones.

Regarding North Korea's potential deployment of additional troops to Russia, President Zelenskyy had indicated since June of this year that preparations were underway in the western Voronezh region, which borders Ukraine, to receive them.

So,  President Zelenskyy says that North Korea has decided to send 30,000 to 50,000 soldiers to Russia. He is calling for support for Ukraine. Apparently, Russia needs more troops to win the war.

人工智能剛剛創造出自然界中不存在的病毒（2/2）

Recently The New York Times reported the following:

This A.I. Just Created Viruses Not Found in Nature (2/2)

Scientists trained artificial intelligence on libraries of DNA and then asked the model to create recipes for viral genomes. Sixteen of them were viable, yielding new viruses.

The NYT - By Carl Zimmer - Carl Zimmer covers news about science for The Times and writes the Origins column.

Aug. 6, 2026

(continue)

They made this choice in part because scientists know Phi X-174 intimately, having studied it for close to a century. And because it’s a bacteriophage that infects only E. coli, they knew that viruses similar to it would be safe.

Once Evo got familiar with the genomes of Phi X-174 and its kin, the researchers prompted the model to write new versions of its own. Evo generated 700,000 potential versions; the scientists pursued only the ones that looked as if they had the best odds of succeeding.

They ended up making DNA molecules from 285 of Evo’s suggested sequences. When those genomes were ready to test, Mr. King and his colleagues inserted them into bacteria, which they spread across petri dishes.

In many of the dishes, the microbes grew peacefully — Evo’s genomes had failed. But in one dish, researchers noticed clear dots appearing in the cloudy film of bacteria, the telltale sign of multiplying viruses.

The DNA that the scientists had inserted into the bacteria had produced protein shells that contained viral genes. The viruses burst out of the cells, leaving behind the ruptured husks of their dead hosts.

As Mr. King and his colleagues tested more genomes, they saw more clear dots. All told, they discovered that 16 of Evo’s genomes produced viable new viruses.

They proved to be as resilient as natural ones. In fact, some multiplied faster than Phi X-174. “They’re not just sickly versions of stuff that already exists,” said Oliver Crook, a protein chemist at the University of Oxford who was not involved in the new study.

Dr. Crook cautioned that Evo’s viruses were not radically new creations. They tend to be very similar to natural species, relying on the same underlying biology.

Scientists will have to run more experiments to find out if Evo has the same success rate if it’s trained on other groups of viruses. If so, Dr. Crook expected that scientists might find some A.I.-generated viruses that could become useful tools for medicine and biotechnology.

“A lot of our science rests on viruses as technology,” he said. To treat people with genetic disorders, for example, doctors will load genes into viruses, which deliver them into cells.

But along with this hope, Evo’s initial success has raised concerns that A.I. could be used to create deadly pathogens. “You could say, ‘Hey, genomic language model, make me an influenza genome that is modified to be more transmissible or to be more lethal,’” Dr. Hanke speculated.

The potential dangers were already on the minds of the researchers as they trained Evo. They did not provide the model with data about viruses that infect humans, and they excluded similar viruses that infect other animals, plants and fungi.

As a result, Evo can’t generate genomes of viruses that could threaten humans. “We just wanted to be extra careful,” said Brian Hie, a computational biologist at Stanford University and an author of the new study.

Dr. Hanke credited Dr. Hie and his colleagues for taking that precaution. “I think that’s quite commendable,” he said, “because they don’t get any guidance from anywhere on what they should be doing.”

Last week, Dr. Hanke noted, the National Institutes of Health rolled out a new policy for stopping high-risk life science research. The policy would bar scientists from experiments that would make biological agents more harmful.

But computer-based research — such as generating virus DNA with A.I. — “is not prohibited by this policy unless it involves an entity of concern,” the agency said in a statement.

It’s easy to tell if a natural virus like smallpox is an entity of concern. But Dr. Hanke said there’s no consensus on judging the possible danger of a virus made by an A.I. model.

“What is the risk of what I’ve never seen before?” he asked.

Translation

人工智能剛剛創造出自然界中不存在的病毒（2/2）

科學家利用DNA庫訓練人工智能，然後讓模型產生病毒基因組的「配方」。其中16個配方成功生成了新的病毒

（繼續）

他們之所以選擇這種病毒，部分原因是科學家對Phi X-174噬菌體非常了解，已經研究了近一個世紀。而且由於它是一種只感染大腸桿菌的噬菌體，科學家知道與其類似的病毒是安全的。

一旦 Evo 熟悉了Phi​​ X-174及其近親的基因組，研究人員就讓模型生成自己的新版本。 Evo產生了70萬個潛在版本；科學家只選擇了那些看起來最有可能成功的版本來研究。

他們最終利用Evo提供的285個序列建構了DNA分子。當這些基因組準備好後，King先生和他的同事將它們植入細菌，並鋪開在培養皿中。

在許多培養皿中，微生物生長良 - 但Evo的基因組未能成功複製。但在一個培養皿中，研究人員注意到細菌的渾濁薄膜上出現了清晰的斑點，這是有病毒在繁殖中的明顯跡象。

被科學家植入細菌的DNA產生了含有病毒基因的蛋白質外殼。病毒從細胞中破殼而出，留下宿主細胞破裂的殘骸。

隨著King先生和他的同事測試更多基因組，他們觀察到了更多清晰的斑點。最終，他們發現Evo提供的16個基因組產生了可行的新病毒。

這些病毒與天然病毒一樣能迅速適應環境變化。事實上，有些病毒的繁殖速度甚至超過了Phi X-174。並未參與這項新研究的牛津大學蛋白質化學家Oliver Crook說道: 「它們並非只是現有病毒的弱態版本」。

Crook博士提醒說，Evo 生成的病毒並非全新物種。它們往往與天然病毒非常相似，依賴相同的生物學基礎。

科學家需要進行更多實驗，以確定如果 Evo 使用其他病毒群進行訓練，其成功率是否相同。如果成功，Crook博士預計科學家可能會發現一些人工智能生成的病毒，這些病毒有望成為醫學上和生物技術領域的有用工具。

他說:「我們的許多科學研究在技術上都依賴病毒」。例如，為了治療遺傳性疾病患者，醫生會將基因加載到病毒中，然後由它導入細胞。

然而，在帶來希望的同時，Evo 的初步成功也引發了人們的擔憂：人工智能可能會被用於製造致命病原體。 Hanke博士推測地説: 「你可以說，『嘿，基因組語言模型，幫我產生一個經過改造的流感病毒基因組，使其更具傳染性或更致命』」。

研究人員在訓練Evo模型時就已經考慮到了潛在的危險。他們沒有向模型提供感染人類的病毒數據，並且排除了類似能感染其他動物、植物和真菌的病毒。

因此，Evo無法產生可能威脅人類的病毒基因組。史丹佛大學計算生物學家、這項新研究的作者之一Brian Hie說:「我們只是想格外謹慎」。

Hanke博士讚揚了Hie博士及其同事採取的這種預防措施。 他說:“我認為這非常值得稱讚”，“因為他們沒有得到任何關於應該做什麼的指導。”

上週Hanke博士指出，美國國立衛生研究院推出了一項新的政策，旨在停止高風險的生命科學研究。這項政策將禁止科學家進行任何可能使生物物質更具危害性的實驗。

但該機構在聲明中表示，基於電腦的研究 - 例如利用人工智能產生病毒DNA -「並不在此政策的禁止範圍之內，除非它涉及令人擔憂的實體」。

判斷像天花這樣的天然病毒是否令人擔憂是很容易。但Hanke博士表示，對於如何判斷人工智能模型產生的病毒可能存在的危險，目前尚無共識。

他問:「我從怎會知道從未見過的事物會帶來什麼風險？」。

              So, for the first time, scientists have used artificial intelligence to create new kinds of viruses, raising hopes for medical advances while also raising the disturbing possibility that the technology could someday be used to invent dangerous pathogens. DNA has its own rules of grammar and biologists have uncovered some of nature’s grammatical rules, but many remain a mystery. In the present case, Evo the A.I. had scanned about nine trillion nucleotides. Evo eventually recognized patterns common across the tree of life and used them to generate blueprints for new genes encoding proteins that could perform specific jobs. Apparently, we should be very careful in doing these kinds of researches because we do not know what is the risk of what we’ve never seen before.

Note:

1. A bacteriophage (噬菌體), also known informally as a phage, is a virus that infects and replicates within bacteria. Bacteriophages are composed of proteins that encapsulate a DNA or RNA genome, and may have structures that are either simple or elaborate. Their genomes may encode as few as four genes (e.g. MS2) and as many as hundreds of genes. Phages replicate within the bacterium following the injection of their genome into its cytoplasm. Bacteriophages are among the most common and diverse entities in the biosphere. Bacteriophages are ubiquitous viruses, found wherever bacteria exist. (Wikipedia)

人工智能剛剛創造出自然界中不存在的病毒（1/2）

Recently The New York Times reported the following:

(Source: The NYT)

This A.I. Just Created Viruses Not Found in Nature (1/2)

Scientists trained artificial intelligence on libraries of DNA and then asked the model to create recipes for viral genomes. Sixteen of them were viable, yielding new viruses.

The NYT - By Carl Zimmer - Carl Zimmer covers news about science for The Times and writes the Origins column.

Aug. 6, 2026

For the first time, scientists have used artificial intelligence to create new kinds of viruses, raising hopes for medical advances while also raising the disturbing possibility that the technology could someday be used to invent dangerous pathogens.

Synthesizing viruses from scratch is hardly new. Researchers long ago learned how to manufacture viral genomes; they are used to investigate antiviral drugs and vaccines, as well as to learn how viruses work.

But the new study, published Thursday in the journal Science, goes well beyond duplicating viral genes. Scientists at Stanford University and the Arc Institute, a research organization in Palo Alto, Calif., taught A.I. to recognize patterns of DNA structure in nature, and then to use that data to write recipes for entirely new viruses.

The researchers followed those recipes to create DNA molecules, which they inserted into bacteria. The modified bacteria then produced viruses never seen in nature. The viruses were able to infect other bacteria, demonstrating that they were viable.

“This is an important milestone,” said Patrick Cai, a synthetic biologist at the University of Manchester, who was not involved in the study.

The viruses dreamed up by A.I. do not pose a threat to humans, because they are all similar to a naturally occurring virus called Phi X-174, which can infect only bacteria.

But the new study adds to growing worries that artificial intelligence might enable the creation of a new generation of biological weapons, from deadly poisons to unstoppable pandemics.

Dr. Moritz Hanke, a fellow at the Johns Hopkins Center for Health Security who was not involved in the new study, said governments and scientific organizations have been slow to develop guardrails that could block the creation of a deadly virus — even as the science races ahead.

“There’s just a huge disconnect,” he said.

The authors of the study relied on an A.I. model called Evo, which is similar in some ways to ChatGPT, made by OpenAI. 

ChatGPT answers questions by stringing together words that seem likely to follow one another. It can do this thanks to years of training on vast amounts of text gathered from the internet, books and other sources.

DNA is strikingly similar to a book in some ways: a string of molecular building blocks, known as nucleotides, arrayed like letters in a line of text. A gene consists of hundreds of nucleotides drawn from a four-letter alphabet: A, C, G and T. The sequence encodes the instructions for building proteins and other molecules.

DNA has its own rules of grammar, and if a sequence violates them, the result is biological gibberish. Biologists have uncovered some of nature’s grammatical rules, but many remain a mystery.

The researchers wondered if Evo could pick up these rules on its own. Instead of training it on text, they trained it on genetic sequences drawn from millions of animals, plants, microbes and viruses. All told, Evo scanned about nine trillion nucleotides.

Evo eventually recognized patterns common across the tree of life and used them to generate blueprints for new genes encoding proteins that could perform specific jobs. These results led the team to wonder if Evo could master not just single genes but also an entire genome.

As the A.I. would be able to handle only small genomes at first, the scientists decided to try to make viruses. While a human genome contains over three billion nucleotides, many viruses have genomes just a few thousand nucleotides long.

“It just felt like the obvious next step,” said Samuel King, a graduate student at Stanford University and an author of the new study. He and his colleagues gave Evo another round of training, this time on the 11 genes of Phi X-174 and about 15,000 of its closest relatives.

(to be continued)

Translation

人工智能剛剛創造出自然界中不存在的病毒（1/2）

科學家利用DNA庫訓練人工智能，然後讓模型創建病毒基因組的合成方案。其中16個方案成功，產生了新的病毒

科學家首次利用人工智能創造了新型病毒，為醫學進步帶來了希望，同時也引發了一個令人不安的可能性：這項技術未來可能被用於製造危險的病原體。

從零開始合成病毒並非新鮮事。研究人員很久以前就掌握瞭如何製造病毒基因組；他們利用病毒基因組來研究抗病毒藥物和疫苗，以及了解病毒的運作機制。

但這項週四發表在《科學》雜誌上的新研究，遠遠超出了複製病毒基因的範疇。史丹佛大學和位於加州Palo Alto的Arc研究所的科學家訓練人工智能辨識自然界中DNA結構的模式，並利用這些資料編寫全新病毒的「配方」。

研究人員依照這些「配方」製造DNA分子，並將其插入細菌體內。改造後的細菌隨後產生了自然界中從未出現過的病毒。這些病毒能夠感染其他細菌，證明它們是可行的。

並未參與這項研究的曼徹斯特大學的合成生物學家Patrick Cai說道:「這是一個重要的里程碑」。

人工智能製造的這些病毒不會對人類構成威脅，因為它們都與一種名為Phi X-174的天然病毒相似，而Phi X-174只能感染細菌。

但這項新研究加劇了人們日益增長的擔憂：人工智能可能被用於製造新一代生物武器，從致命毒藥到無法控制的流行病。

約翰霍普金斯大學健康安全中心的研究員Moritz Hanke博士並未參與這項新研究。他表示，儘管科學發展日新月異，但各國政府和科研機構在制定出能夠阻止致命病毒產生的防止措施方面卻進展緩慢。

他說:「這之間存在著巨大的脫節」。

研究的作者使用了名為Evo的人工智能模型，它在某些方面與OpenAI開發的ChatGPT類似。

ChatGPT透過將看似可能連續出現的字詞串連起來去回答問題。它之所以能夠做到這一點，是因為它經過多年對從互聯網、書籍和其他來源收集的大量文本進行訓練。

DNA在某些方面與書籍驚人地相似：它們都是由被稱為核苷酸 (nucleotides) 的分子構件組成的，排列方式如同文本中的字母。基因由數百個核苷酸組成，這些核苷酸取自字母表：A、C、G 和 T。此序列編碼構成對蛋白質和其他分子的指令。

DNA 有其自身的語法規則，如果序列違反這些規則，就會產生生物亂碼。生物學家已經揭示了一些自然界的語法規則，但許多規則仍然是個謎。

研究人員想知道 Evo 是否能夠自主學習這些規則。他們沒有用文字訓練 Evo，而是用來自數百萬種動物、植物、微生物和病毒的基因序列來訓練。 Evo 總共掃描了大約 9 兆個核苷酸。

Evo 最終識別出了生命之樹中常見的模式，並利用這些模式產生了編碼能夠執行特定功能的蛋白質的新基因藍圖。這些結果讓研究團隊開始思考，Evo 是否不僅能掌握單一基因，還能掌握整個基因組。

由於人工智能最初只能處理小型基因組，科學家決定嘗試建構病毒。人類基因組包含超過30億個核苷酸，而許多病毒的基因組只有數千個核苷酸長。

史丹佛大學研究生、這項新研究的作者之一 Samuel King 說:「這感覺就像是順理成章的下一步」。他和同事們對Evo進行了新一輪的訓練，這次對象是Phi X-174的11個基因以及大約15,000個與其親緣關係最近的基因。

 （待續）

中國的人工智能正在非洲迅速發展，這應該引起矽谷的擔憂（3/3）

Recently The New York Times reported the following:

(Source: The NYT)

China’s A.I. Is Surging Across Africa. That Should Worry Silicon Valley (3/3)

In African tech hubs, developers are picking China’s cheap, freely available artificial intelligence models over more powerful U.S. ones.

The NYT - By Paul MozurAdam Satariano and Aaron Krolik (Paul Mozur and Adam Satariano reported from Nairobi, Kenya, and Aaron Krolik from New York.)

Aug. 5, 2026

(continued from part 2)

Wake Up Call

In a Nairobi high-rise in June, a top Kenyan civil servant for technology said he was surprised when Anthropic pulled access to Fable, its powerful model, at the Trump administration’s behest.

It was a “wake up call” to the world, said John Tanui, the principal secretary in Kenya’s Ministry of Information, Communications and the Digital Economy. The abrupt move was a warning that relying on U.S. models could pose risks, he said.

A former Huawei employee who once led the Konza Technopolis project, Mr. Tanui said China and the United States had pressed Kenya to pick a side. He intends to take from both.

“We don’t lean on West or East,” he said.

Being in the middle is tricky. Developers are building systems controlled elsewhere, which means the technology can be yanked, altered or re-priced without warning. Some worry about dependence on the United States. Others do not trust China.

Mr. Budhabhatti of Craft Silicon said he was considering Huawei’s offer of generous subsidies, but was also cautious. Many of his customers are international banks that could demand their software not use Chinese hardware.

Chinese models, too, have not always been reliable. Shikoh Gitau, the founder of the Nairobi start-up Qhala, recalled when DeepSeek went dark in China before the country’s college entrance exams in 2024. It was to deter cheating, but it also disrupted African businesses like hers. Now American models could vanish because of a White House decision.

“At least with China’s models, I can download it on my computer and run it,” she said.

Others have been burned by A.I. politics. Mr. Mwebaze’s Ugandan chatbot service, Sunflower, ran into trouble when a researcher at China Media Project, an independent media research group, asked it about China. Sunflower described China as a democracy and sidestepped questions about the country’s economic and environmental impacts in Uganda.

Chinese state media praised Sunflower, and Huawei contacted Mr. Mwebaze about using its cloud services.

Mr. Mwebaze, whose nonprofit organization, Sunbird AI, is dedicated to A.I. for social good, was unnerved by the experience. He is now using Google’s open-source model, Gemma, to create a product aimed at smartphone users.

“We need to make sure we’re not on the wrong side of geopolitics,” he said.

America Strikes Back

For all of China’s momentum, much of the A.I. money in Kenya still flows to U.S. companies.

Everyday users are on U.S. systems like ChatGPT and Claude. One 2025 survey found that 42 percent of Kenya’s 23 million internet users had used ChatGPT in the previous month, among the highest rate in the world.

Even Chinese models earn Americans money. Kenyan companies often run Chinese open-source models on cloud services from Amazon and Microsoft, so the U.S. giants make money from their data centers that deliver the technology to customers.

The U.S. advantage was clear in a concrete building in Nairobi’s industrial district, where binders of paper purchase orders lined dark offices for Chandaria Industries, one of Kenya’s largest makers of toilet paper, soap and home goods.

For decades, orders arrived from supermarkets by email or on paper, and clerks keyed them in line by line. An order could take hours to process.

Chandaria recently hired Sanifu, a Kenyan start-up that uses A.I. to streamline business processes, for help. Sanifu uses OpenAI’s models for precision and reliability.

Sanifu’s digital system based on those U.S. models has transformed Chandaria. “Where an order took about two hours to process, that now takes less than 10 minutes,” said Neer Chandaria, who helps run the family firm. Four clerks who once typed in orders now audit the A.I.’s work.

But Sanifu is being courted too. On LinkedIn this year, a Singaporean employee of the Chinese internet firm ByteDance offered Sanifu’s founders access to its A.I. model.

ByteDance’s model was not accurate enough, but the courtship made an impression, said Bernard Momanyi Nyagaka, a co-founder of Sanifu. Nothing like it comes from the American labs.

And Chinese models keep improving, Mr. Nyagaka said.

“If you look at how fast they’re catching up with the American models, it’s really, really fast,” he said.

Translation

中國的人工智能正在非洲迅速發展，這應該引起矽谷的擔憂（3/3）

在非洲的科技中心，開發者們正在選擇中國廉價且免費的人工智能模型，而不是功能更強大的美國模型

（接續第二部分）

警醒

今年6月，在內羅比的一棟高層建築裡，一位肯亞高級技術公務員表示，他對Anthropic公司應朗特普政府的要求撤回其強大的人工智能模式Fable的存取權限感到驚訝。

肯亞資訊、通訊和數位經濟部常務秘書John Tanui表示，這給世界一個「警醒」。他說，這項突然舉動是在警告人們，依賴美國模型可能會帶來風險。

曾任華為員工、並領導Konza Technopoli科技城計劃的Tanui先生表示，中國和美國都曾施壓肯亞選邊站隊。他打算兼顧雙方。

他說「我們不偏袒西方，也不偏袒東方」。

居於中立位置並非易事。開發者正在建立由其他國家控制的系統，這意味著技術可能在毫無預警的情況下被撤回、修改或重新定價。有些人擔心過度依賴美國，而有些人則不信任中國。

Craft Silicon公司的Budhabhatti先生表示，他正在考慮華為提供的豐厚補貼，但也保持謹慎。他的許多客戶都是國際銀行，這些銀行可能會要求其軟件不得使用在中國的硬件。

此外，中國模式也並非總是可靠。內羅比初創公司Qhala的創辦人Shikoh Gitau回憶起，在2024年中國高考前夕，DeepSeek在中國停止營運。此舉旨在防止作弊，但也擾亂了像她這樣的非洲企業。如今，由於白宮的一項決定，美國的人工智能模型可能就會消失。

她說:「至少我可以下載中國的型號到電腦上運行」。

其他人則因人工智能政治而蒙受損失。Mwebaze先生的叫做「向日葵」（Sunflower）的烏干達 聊天機器人服務就遇到了麻煩。當時，獨立媒體研究機構「中國傳媒計劃」（China Media Project）的一位研究員向「向日葵」詢問了有關中國的問題。 「向日葵」將中國描述為一個民主國家，並迴避了有關中國對烏干達經濟和環境影響的問題。

中國官方媒體對「向日葵」大加讚揚，華為也聯繫了Mwebaze先生，希望他使用其雲端服務。

Mwebaze先生的非營利組織「太陽鳥人工智能」（Sunbird AI）致力於利用人工智能造福社會，這次經歷讓他感到不安。他現在正在使用Google的開源模型Gemma開發一款面向智能型手機用戶的產品。

他說:「我們需要確保自己不會站在地緣政治的錯誤一邊」。

美國反擊

儘管中國在人工智領域勢頭強勁，但肯亞的大部分人工智能資金仍流向美國公司。

肯亞的一般用戶都在使用ChatGPT和Claude等美國系統。一項2025年的調查發現，肯亞2,300萬網路用戶中有42%的人在上個月使用過ChatGPT，這比例位居世界前列。

即使是中國的人工智能模型也賺美國錢。肯亞公司經常在亞馬遜和微軟的雲端服務上運行中國的開源模型，這些美國巨頭可以從其數據中心負責向客戶提供技術服務中獲利。

在內羅比工業區的一棟混凝土建築裡，美國的優勢顯而易見。肯亞最大的衛生紙、肥皂和家居用品生產商之一 - 錢德里亞工業公司（Chandaria Industries）昏暗的辦公室裡，一疊疊紙質採購訂單整齊地擺放著。

幾十年來，超市的訂單都是透過電子郵件或紙本形式送達，店員需要逐行輸入。處理一份訂單可能需要幾個小時。

Chandaria 最近聘請了肯亞新創公司 Sanifu 來協助其業務流程優化。 Sanifu 利用人工智能技術簡化業務流程，並使用 OpenAI 的模型來確保精準性和可靠性。

Sanifu 基於這些美國模型的數位化系統徹底改變了 Chandaria 的營運模式。協助管理家族企業的 Neer Chandaria 說: 「以前處理訂單需要大約兩個小時，現在不到 10 分鐘就能完成」。過去負責輸入訂單的四名職員現在負責審核人工智能的處理結果。

Sanifu 也受到了其他公司的青睞。今年，一位來自中國網絡公司字節跳動的新加坡員工在 LinkedIn 上向 Sanifu 的創辦人提供了其人工智能模式的使用權限。

Sanifu 的共同創辦人 Bernard Momanyi Nyagaka 表示，字節跳動的模型雖然不夠精確，但這次青睞卻給他們留下了深刻的印象。美國實驗室從未有類似的產品。

Nyagaka 先生也表示，中國的模型也在不斷改進。

他說:「如果你看看他們追趕美國模型的速度，你會發現是真的非常非常快」。

              So, the A.I. model from the Chinese internet giant Alibaba handles Uganda’s dozens of languages better than anything from Meta or Google. It is inexpensive, and could be customized. In Kenya, entrepreneurs are using the models to streamline legal and business services. China and the United States have pressed Kenya to pick a side.  Some users worry about dependence on the United States. Others do not trust China. Apparently, Africa is a place where China and the US will continue fighting for the AI market.

中國的人工智能正在非洲迅速崛起，這應該引起矽谷的擔憂（2/3）

Recently The New York Times reported the following:

(Source: The NYT)

China’s A.I. Is Surging Across Africa. That Should Worry Silicon Valley (2/3)

In African tech hubs, developers are picking China’s cheap, freely available artificial intelligence models over more powerful U.S. ones.

The NYT - By Paul MozurAdam Satariano and Aaron Krolik (Paul Mozur and Adam Satariano reported from Nairobi, Kenya, and Aaron Krolik from New York.)

Aug. 5, 2026

(continue from part 1)

Yet Africa’s experience shows the global A.I. competition is now a two-horse race.

Kamal Budhabhatti, the chief executive of Craft Silicon, a Nairobi banking software company, feels the pull of both sides. He recently rebuilt his firm’s tech systems with Claude Code, Anthropic’s programming tool. A 10-person team finished in half the time a project that once took 100 engineers. The bill was high, he said, but the speed was worth it.

Huawei, the Chinese tech giant, also came calling with “some very unbelievable incentives” to switch to Chinese systems, he said, including a year of free computing and a trip to the company’s headquarters.

“It’s very hard to say no,” Mr. Budhabhatti said.

The China Model

About 50 miles southeast of Nairobi, one of Kenya’s biggest collaborations with Chinese technology sits half-finished on the savanna.

Konza Technopolis, a planned city, was announced 18 years ago as Kenya’s answer to Silicon Valley. Hopefully branded the Silicon Savannah, it is now a grid of empty boulevards and skeletal buildings, watched over by a powerful Chinese surveillance system.

It would look like a failed experiment but for one slate-gray building behind an electric fence. Financed by Chinese loans and built by Huawei, the site is a data center that runs computing for Kenya’s government.

For two decades, Chinese firms have wired Africa’s telecommunications networks and paved its roads. M-Pesa, Kenya’s celebrated mobile-money system, runs on Huawei technology.

These tech ties gave China a foothold in the continent, which it is using to sell A.I. One project advertised at Konza will use technology from the Chinese A.I. firm DeepSeek to combat telecom fraud.

Locked out of advanced chips because of U.S. export controls, Chinese companies could not outspend Silicon Valley. So they gave their A.I. models away as open source, hoping to pull people from U.S. rivals.

At first, it seemed a losing bet. The United States dominated open-source A.I. Meta offered a model called Llama that developers used worldwide. But after some stumbles, Mark Zuckerberg, Meta’s chief executive, turned more to closed models.

“It left a huge hole,” said Lucas Atkins, co-founder of Arcee A.I., an American start-up that builds open models. “It allowed the Chinese to explode into the vacuum.”

Breakthroughs came quickly. In December 2024, a DeepSeek A.I. model matched the best models at a fraction of the cost. Last month, the Chinese A.I. lab Moonshot AI released a model with coding abilities approaching the leading American systems.

In Kenya, developers embraced Chinese models.

Michael Michie builds A.I. systems for Kenyan banks and government agencies at EverseTech. His customers are price sensitive, so he makes almost everything on Chinese open-source models. Businesses want the cheapest way to use the technology, not the most advanced, he said.

“I don’t think people should worry so much about who built it,” he said. “The focus should be, does it deliver the capability you need?”

Sentai Simons, a start-up founder in Nairobi, downloaded Chinese models to build a product that turned paper legal records into searchable databases. He customized the system with about one million legal files.

His product, JibuDocs, ended up about 140 gigabytes in size — smaller than many iPhones — and cost about $25,000 to create and maintain. Using a service like Anthropic’s Claude, he said, would have cost more than $1 million.

“It’s absolutely the difference between having a business and not having a business,” he said.

Many developers said Chinese companies treated them like a priority, while U.S. firms were hardly present.

Ahmet Acar, a former principal adviser at Amazon Web Services who lives in Nairobi, said joining a developer program from Anthropic required a lengthy screening. Alibaba’s took minutes.

“With the Chinese, it was like: ‘Hey, are you able to do this? Are you located here? All right, man, go have fun,’” he said.

(to be continued in part 3)

Translation

中國的人工智能正在非洲迅速崛起，這應該引起矽谷的擔憂（2/3）

在非洲的科技中心，開發者們正在選擇中國廉價且免費的人工智能模型，而不是功能更強大的美國模型

（接上文）

然而，非洲的經驗表明，全球人工智能競爭如今已演變為一場雙雄爭霸。

內羅比銀行軟件公司Craft Silicon的執行長Kamal Budhabhatti感受到了雙方的拉鋸。他最近使用Anthropic公司的程式設計工具Claude Code重建了公司的技術系統。一個10人的團隊僅用一半的時間就完成了原本需要100名工程師才能完成的專案。他說，費用很高，但速度快是得值的。

他還說，中國科技巨頭華為也主動聯繫他們，提供了“一些令人難以置信的優惠”，鼓勵他們改用中國系統，包括一年的免費運算服務和一次參觀華為總部的機會。

Budhabhatti先生說: 「這是很難拒絕的」。

中國模式

在內羅比東南約50英里處，肯亞與中國科技公司合作的最大計劃之一，如今卻荒廢在只有少量樹木的草原之上。

Konza Technopolis是個規劃中的城市，18年前宣佈建設，旨在成為肯亞的「矽谷」。它曾被寄予厚望，希望被譽為 “矽谷草原” ，但如今卻只是一片空蕩蕩的林蔭大道和只有骨架的建築，被強大的中國監控系統監視著。

這看起來像是一項失敗的實驗，但電動圍欄後面有一棟石板灰色的建築物。這地方是個由中國貸款融資的數據中心，由華為承建，為肯亞政府提供運算服務。

二十年來，中國企業為非洲的電信網絡和道路建設做出了巨大貢獻。肯亞著名的行動支付系統M-Pesa就採用了華為的技術。

這些技術聯繫使中國在非洲大陸站穩了腳跟，並以此為據點推銷人工智能技術。在Konza大會上展示的一個項目將利用中國人工智能公司DeepSeek的技術來打擊電信詐欺。

由於美國的出口管制，中國企業無法取得先進晶片，也無法在資金上與矽谷匹敵。因此，他們將人工智能模式開源，希望以此吸引美國競爭對手的人才。

起初，這似乎是一場注定失敗的賭博。美國在開源人工智能領域佔據主導地位。 Meta公司提供了一個名為Llama的模型，並被世界各地的開發者使用。但在經歷了一些挫折之後，Meta的執行長朱克伯格開始更多地轉向封閉式模型。

致力於建立開源模型的美國初創公司 Arcee A.I. 的聯合創始人Lucas Atkins 說: 「這留下了一個巨大的空檔」;「這讓中國企業得以迅速填補這一真空」。

突破性進展來得很快。 2024 年 12 月，DeepSeek A.I. 的一款模型能以一小部分的成本擁有可與最佳型號模型匹配的效果。上個月，中國人工智能實驗室 Moonshot AI 發佈了一款編碼能力接近美國領先系統的模型。DeepSeek A.I. 模型以一小部分的成本媲美最優秀的模型

在肯亞，開發者們接受了中國模型。

Michael Michie 在 EverseTech 公司為肯亞的銀行和政府機構建立人工智能系統。他的客戶對價格非常敏感，因此他幾乎所有系統都基於中國的開源模型。他說，企業想要的是最經濟的技術使用方式，而不是最先進的技術。

他說:「我認為人們不應該過分在意是誰開發的」 ;「重點應該是，它是否能提供你所需的功能？」。

內羅比的創業家Sentai Simons借鑒了中國的模式，開發了一款可以將紙本法律記錄轉換為可搜尋資料庫的產品。他用大約一百萬份法律文件對系統進行了度身定做。

他的產品 JibuDocs 最終大小約為 140 GB - 比許多 iPhone 都小 - 開發和維護成本約為 25,000 美元。他說，如果使用像 Anthropic 的 Claude 這樣的服務，成本將超過 100 萬美元。

他說:「「這絕對是有生意和沒生意之間的差別」。

很多開發者說中國公司把他們當成優先考慮的對象，而美國公司幾乎不存在。

居住在內羅比的亞馬遜網路服務（AWS）前首席顧問Ahmet Acar表示，去加入 Anthropic 的開發者的計劃需要經過漫長的篩選，而阿里巴巴的篩選只需幾分鐘。

他說:「跟中國人合作時，感覺就像是：『喂，你能做到嗎？你在本地嗎？好吧，兄弟，一齊嚟喇』」。

 （待續，見第三部分）

中國的人工智能正在非洲迅速崛起，這應該引起矽谷的擔憂 (1/3)

Recently The New York Times reported the following:

(Source: The NYT)

China’s A.I. Is Surging Across Africa. That Should Worry Silicon Valley (1/3)

In African tech hubs, developers are picking China’s cheap, freely available artificial intelligence models over more powerful U.S. ones.

The NYT - By Paul MozurAdam Satariano and Aaron Krolik (Paul Mozur and Adam Satariano reported from Nairobi, Kenya, and Aaron Krolik from New York.)

Aug. 5, 2026

When Ernest Mwebaze, a tech developer in Uganda, was building an artificial intelligence system last year tailored for his country’s many languages, he tested American and Chinese tools to see which one could help.

China’s won.

The A.I. model from the Chinese internet giant Alibaba handled Uganda’s dozens of languages better than anything from Meta or Google, Mr. Mwebaze said. It was also inexpensive, and he could customize the model with his own data.

“We want to build things as cheap as possible, yet have them work really well,” said Mr. Mwebaze, a former research scientist at Google. His system, called Sunflower, is now used across Uganda, including by farmers to receive weather information and crop advice in local dialects.

Mr. Mwebaze, 47, is one of thousands of developers across Africa who have turned to Chinese A.I. models in the past year. In Kenya, entrepreneurs are using the models to streamline legal and business services. In Nigeria, they have made educational tools that teach high school students. In Ghana, developers are building local chatbots.

China’s A.I. is surging, especially in developing countries, as people look for the best possible system at the lowest possible cost. Unlike models made by the leading American A.I. companies OpenAI and Anthropic — which are closed and charge fees — the Chinese systems are publicly available to download and modify without payment or approval.

Chinese “open-source” models account for roughly half of total A.I. use on OpenRouter, a service with 400 A.I. models for users to choose from, up from less than 25 percent a year ago, according to a New York Times analysis of data from eight million customers. Chinese models are also 19 of the 25 most downloaded open-source systems on Hugging Face, another A.I. database.

In more than two dozen interviews across Kenya, Uganda and other countries, developers, policymakers and executives described a rapidly changing A.I. market where cost, computing resources, data control and customization mattered more than where a model was made. Often they were not looking for bleeding edge A.I. They just wanted something that worked well enough on limited resources.

Using Chinese models was like owning a house while using U.S. models was like renting one, they added. Rather than spending millions to train a model from scratch, developers can download a Chinese system and build an entirely new product on top of it, paying mainly for servers. In contrast, U.S. models are a one-size-fits-all technology, needing subscriptions and other fees depending on the job.

“Why use an expensive Ferrari to do the school run when a Toyota hatchback can do the same?” said Moses Kemibaro, a Nairobi entrepreneur who runs a digital marketing agency, Dotsavvy, adding that Chinese models can be up to 90 percent less expensive when including costs like computing infrastructure.

Chinese officials hope their gains in Africa are a preview of things to come. Last month, powerful new Chinese A.I. systems rattled Wall Street and Silicon Valley, stoking fears in Washington that U.S. tech dominance could be undercut.

Xi Jinping, China’s leader, is using A.I. as a form of soft power. At a July conference in Shanghai, he cast Chinese models as more reliable and cheaper and warned that A.I.’s benefits must be shared or they would create “new historical injustices.” Kenya, Ethiopia, South Africa and seven other African countries signed an A.I. pact with China at the event to promote international cooperation.

Chinese firms are also courting African developers with free computing and hands-on engineering help, developers said. Many Chinese-made smartphones in Africa come with Chinese A.I. tools preinstalled.

At the same time, China’s A.I. industry faces challenges. Companies have struggled to make money from their users and face security questions about data and links to Beijing. Many African developers still pay for American A.I. models, especially for technical tasks like coding.

(to be continued in part 2)

Translation

中國的人工智能正在非洲迅速崛起，這應該引起矽谷的擔憂  (1/3)

在非洲的科技中心，開發者們正在選擇中國廉價且免費的人工智能模型，而不是功能更強大的美國模型

去年，烏干達的一位科技開發者Ernest Mwebaze在建立一​​個針對該國多種語言的人工智能系統時，測試了美國和中國的工具，看看哪個更好用。

中國的模型勝出了。

Mwebaze先生表示，來自中國網絡巨頭阿里巴巴的人工智能模式在處理烏干達數十種語言方面，比Meta或Google的任何產品都做得更好。而且，它價格低廉，它還可以使用自己的數據為客人度身定做。

谷歌前研究科學家nMwebazen先生說道:「我們希望以盡可能低的成本打造出性能卓越的產品」。他開發的名為「向日葵」（Sunflower）的系統目前已在烏干達廣泛應用，包括農民用它來獲取當地語言的天氣資訊和作物種植建議。

47歲的 Mwebaze 先生是過去一年轉向中國人工智能模型的數千名非洲開發者之一。在肯亞，企業家們正在利用這些模式簡化法律和商業服務流程。在尼日利亞，他們開發了高中生教學的教育工具。在加納，開發者們正在建造本地化的聊天機器人。

中國的人工智能正在蓬勃發展，尤其是在發展中國家，因為人們都在尋求以盡可能低的成本獲得最佳的系統。與美國領先的人工智能公司OpenAI和Anthropic開發的封閉式收費模型不同，中國的系統可以公開下載和修改，無需付費或獲得批准。

中國的「開源」模型約佔 OpenRouter 的人工智能總量的一半。根據《紐約時報》對八百萬用戶的數據分析顯示，OpenRouter是一項提供400個AI模型供用戶選擇的服務，而中國模型的使用率較一年前的不足25%已大幅上升。此外，在另一個AI資料中心Hugging Face上，在下載量最高的25個開源系統中，有19個是中國模型。

在肯亞、烏干達和其他國家的二十多次訪談中，開發者、政策制定者和企業主管描述了一個快速變化的AI市場，在這個市場中，對於成本、運算資源、資料控制和度身定做方面，相對比模型的出產地更為重要。他們通常不會追求最尖端的AI技術，而只是想要一個能夠在資源有限的情況下運作良好的系統。

他們補充說，使用中國模型就像擁有自己的房子，而使用美國模型就像租房子一樣。開發者無需花費數百萬美元從頭開始訓練模型，只需下載一個中國系統，並在此基礎上建立一個全新的產品，主要費用僅用於伺服器。相較之下，美國模式是一種一機通用的技術，並需要根據具體任務支付服務費和其他費用。

內羅比企業家Moses Kemibaro說道:「為什麼非要駕駛昂貴的法拉利去接送孩子上學，豐田掀背車就能做到喇？」。他經營著一家名為Dotsavvy的提供網絡行銷服務的公司。他還補充說，如果將計算基礎設施等成本考慮在內，中國人工智能模式的價格可以便宜高達90%。

中國官員希望他們在非洲的成就預示著未來發展趨勢。上個月，中國強大的新型人工智能系統震驚了華爾街和矽谷，加劇了華盛頓對美國科技主導地位可能受到削弱的擔憂。

中國領導人習近平正在將人工智能作為一種軟實力手段。在7月於上海舉行的一次會議上，他將中國人工智能模式描述為更可靠、更便宜，並警告說，人工智能的益處必須共享，否則將造成「新的歷史性不公」。肯亞、衣索比亞、南非和其他七個非洲國家與中國簽署了一項人工智能合作協議以促進國際合作。

開發者表示，中國企業也透過提供免費運算資源和實際工程技術支援來吸引非洲開發者。許多在非洲銷售的中國產智能型手機都預先安裝了中國的人工智能工具。

同時，中國的人工智能產業也面臨挑戰。企業難以從用戶身上獲利，並面臨資料安全以及與北京有聯繫等問題的擔憂。許多非洲開發者仍需要付費購買美國的人工智能模型，尤其是在編碼等技術任務方面。

（待續，見第二部份）