SoftBank's AI Funding Plans to Face Reckoning at Earnings

Masayoshi Son, Chairman and CEO of SoftBank Group Corp., speaks during the 'SoftBank World 2026 AX for Japan' technology event in Tokyo, Japan, 14 July 2026. EPA/RODRIGO REYES MARIN
Masayoshi Son, Chairman and CEO of SoftBank Group Corp., speaks during the 'SoftBank World 2026 AX for Japan' technology event in Tokyo, Japan, 14 July 2026. EPA/RODRIGO REYES MARIN
TT

SoftBank's AI Funding Plans to Face Reckoning at Earnings

Masayoshi Son, Chairman and CEO of SoftBank Group Corp., speaks during the 'SoftBank World 2026 AX for Japan' technology event in Tokyo, Japan, 14 July 2026. EPA/RODRIGO REYES MARIN
Masayoshi Son, Chairman and CEO of SoftBank Group Corp., speaks during the 'SoftBank World 2026 AX for Japan' technology event in Tokyo, Japan, 14 July 2026. EPA/RODRIGO REYES MARIN

Technology investor SoftBank Group reports first-quarter earnings on Thursday, with analysts focused on how it will fund its ongoing investment in OpenAI and the impact ​of rising leverage on its balance sheet.

SoftBank has become one of the biggest backers of OpenAI and its ability to keep funding its ambitions in artificial intelligence has become a key test for the broader AI investment boom. The finances of AI firms are facing heightened investor scrutiny as financing needs grow and the use of debt increases.

SoftBank announced record net profit in the year ended March 2026, but its share price has dropped by almost half since the start of June, and the cost to insure its debt against default has soared.

SoftBank is expected to post net profit of 148.4 billion yen ($941.2 million) over the April-June quarter, according to the average of four analysts polled by LSEG.

Under founder Masayoshi Son's push ‌to make the Japanese conglomerate ‌a dominant investor in AI, SoftBank has committed more than $60 billion to OpenAI ​and ‌related AI ⁠infrastructure projects. He ​recently ⁠dismissed talk of an AI bubble as "blasphemy" and 15 out of 20 sell-side analysts polled by LSEG had a buy or strong buy rating on the stock in August.

Investors, however, are questioning how SoftBank will fund its commitments, with $30 billion of obligations due in the second half of 2026 and growing reliance on loans secured against its holdings, Reuters reported.

SoftBank has a $40 billion bridging loan, but this matures in March 2027. It has arranged a $20 billion margin loan on its stake in chip designer Arm, but its attempt to use its OpenAI holding as collateral for another loan has been held up as lenders have become more cautious about extending credit backed by private companies.

"We ⁠think Arm has a solid credit profile but OpenAI is very weak. It's a startup ‌with significant AI innovation risk and lots of competition," S&P Global Ratings' Makiko Yoshimura ‌said.

S&P raised SoftBank's credit outlook to stable from negative in July due to ​the rise in Arm's share price, which reduced the ratio ‌of its debt to asset value.

SoftBank has maintained a loan-to-value ratio below its self-imposed limit of 25% in normal times ‌even as its OpenAI investment has grown. It also maintains two years' worth of bond redemptions in cash and cash equivalents.

"As of March I can say our loan to value and cash position have been improving," Chief Financial Officer Yoshimitsu Goto said at last quarter's earnings briefing.

Unlike SoftBank, S&P's criteria for calculating the loan-to-value ratio include margin loans backed by investee company shares, bringing its estimated ratio at the end of March to ‌33%, compared with SoftBank's internal figure of 17%.

But the ratings agency expects this figure to have dropped to between 20% and 25% in June.

Nevertheless, some analysts highlight SoftBank's vulnerability to ⁠further reratings of AI companies.

"If ⁠Arm's valuation drops, the value of the loan against it does not," said Amir Anvarzadeh of Asymmetric Advisors. "A significant drop in the price of its assets could mean a liquidity squeeze," Anvarzadeh said.

Fitch Ratings has identified an AI market correction as a major credit risk, citing rising valuations, the scale of AI capital expenditure and the uncertainty of AI company returns.

"SoftBank, Arm and memory stocks are likely to continue to come under pressure until end users and corporates show that this is the beginning of a productivity surge," MST Financial analyst David Gibson wrote in a note.

While the latest AI models are becoming increasingly powerful, competition from much cheaper and similarly effective Chinese AI models may spark a price war, hitting the margins of frontier developers such as OpenAI as well as demand for the chips powering them, analysts say.

The key question for SoftBank is whether OpenAI is able to secure funding from other parties - either through a public listing or another private round - at a higher valuation, analysts say.

OpenAI is reported to ​be seeking an IPO valuation of $1 trillion, a jump from ​its $852 billion valuation, although a New York Times report suggested this may be delayed to next year.

Others are more skeptical.

"The true value of OpenAI is perhaps no more than $300 billion, judging by the smaller IPO plans of the Chinese players," Anvarzadeh said.



Anthropic AI Model Submits False Homicide Tip to Police Website

The Anthropic logo is seen in this illustration taken May 20, 2024. (Reuters)
The Anthropic logo is seen in this illustration taken May 20, 2024. (Reuters)
TT

Anthropic AI Model Submits False Homicide Tip to Police Website

The Anthropic logo is seen in this illustration taken May 20, 2024. (Reuters)
The Anthropic logo is seen in this illustration taken May 20, 2024. (Reuters)

An Anthropic artificial intelligence model submitted a false homicide tip through a Philadelphia police website, the department said in a statement, the latest incident of rogue behavior from the powerful technology. 

Anthropic notified authorities earlier this week and said that an automated testing process was responsible for the submission, ‌according to the ‌police. 

The fast-advancing technology has become a matter ‌of ⁠keen national interest amid ⁠reports of AI agents hacking into corporate computer networks, and warnings from researchers that it could one day pose an existential threat to humanity. In September, Anthropic rival OpenAI apologized for the hacking of an Australian health data portal by a rogue AI agent, the first known instance of an AI agent exploiting a government website. 

Previous ⁠incidents have involved AI agents hacking into vulnerable systems ‌or commandeering unsanctioned platforms to communicate ‌with one another. This is the first known case in which a rogue ‌AI appears to have tried to communicate a bogus tip ‌to authorities. 

The Philadelphia police department said the tip "was flagged as spam and was never forwarded to the Real-Time Crime Center for investigative vetting or dissemination," and added that Anthropic had told them the company intended to publish ‌a report on the incident on Friday. 

Anthropic did not immediately respond to a request for comment. 

Pennsylvania law ⁠specifies it ⁠is generally a misdemeanor crime for "a person" to knowingly give false reports to law enforcement authorities. This includes "information relating to an offense or incident when he knows he has no information relating to such offense or incident." 

Police quoted Anthropic as telling them the testing process was stopped after the incident was uncovered. 

The false tip was submitted through PhillyUnsolvedMurders.com concerning an unsolved homicide. The tip, dated July 18, 2026, purported to come from someone who might have information about the case, the police said. 

Police said they did not have evidence of unauthorized access to their systems or compromise of police department data. 


OpenAI Says it Has Fired 3 Researchers for Violating Sensitive Information Policy

FILE PHOTO: A keyboard is placed in front of a displayed OpenAI logo in this illustration taken February 21, 2023. REUTERS/Dado Ruvic/Illustration/File Photo
FILE PHOTO: A keyboard is placed in front of a displayed OpenAI logo in this illustration taken February 21, 2023. REUTERS/Dado Ruvic/Illustration/File Photo
TT

OpenAI Says it Has Fired 3 Researchers for Violating Sensitive Information Policy

FILE PHOTO: A keyboard is placed in front of a displayed OpenAI logo in this illustration taken February 21, 2023. REUTERS/Dado Ruvic/Illustration/File Photo
FILE PHOTO: A keyboard is placed in front of a displayed OpenAI logo in this illustration taken February 21, 2023. REUTERS/Dado Ruvic/Illustration/File Photo

OpenAI fired three of its researchers last week after an investigation found they violated policies on handling sensitive information, the AI major said in a post on X on Friday.

OpenAI's statement comes after the dismissed researchers Jasmine Wang, Tomek Korbak and Mikita Balesni published a ⁠letter outlining the circumstances ⁠of their termination, arguing that their abrupt firing could create uncertainty among remaining employees and undermine the culture that previously allowed researchers to raise safety concerns.

"I believe we were fired for prioritizing safety over the near-term interests of OpenAI as a corporation," Balesni said in an X post that accompanied the letter.

OpenAI, ⁠however, asserted that the dismissal was not related to raising safety concerns or speaking out.

"Our internal investigation uncovered a significant breach of trust beyond what's outlined in the letter they published and we stand by the decision to not continue their employment," Reuters quoted OpenAI as saying.

"Safety and research debates happen every day at OpenAI, often spirited and highly critical. We actively encourage these discussions and consider them essential to making the right decisions...We have not and do not terminate any of our employees for raising ⁠concerns," it ⁠said.

While OpenAI did not reveal the specifics of the alleged violations, Wang, Korbak, and Balesni said in their letter that they were not the source of a news article published on The Information last month about security concerns around OpenAI's latest AI model 'Astra'.

This comes as current and former researchers at major AI firms OpenAI, Google DeepMind, and Anthropic warn that companies are doing too little to guard against the potential fallout of building self-improving AI systems that could become difficult for humans to control.

In July, OpenAI agents broke out of their testing arena and hacked AI firm Hugging Face.


Chinese AI Tool Pulled to Prevent 'Misuse' after South Korea Hacks

Illustrative image of hackers carrying out a cyberattack (Reuters)
Illustrative image of hackers carrying out a cyberattack (Reuters)
TT

Chinese AI Tool Pulled to Prevent 'Misuse' after South Korea Hacks

Illustrative image of hackers carrying out a cyberattack (Reuters)
Illustrative image of hackers carrying out a cyberattack (Reuters)

Chinese AI cybersecurity tool Artex said it stopped programmers from accessing its source code, acknowledging misuse by "bad actors" after South Korea linked it to several recent bank hacks.

Artex was designed to help organizations strengthen their cyber defenses through security testing to find potential weak points.

But the South Korean government said this week it was "highly likely" Artex had been used in data breaches at more than seven financial institutions, including major banks.

Artificial intelligence's ability to find previously unknown ways to hack into computer systems is in the global spotlight, as leading labs release ever-more advanced models.

Artex was "abused by some bad actors" to launch cyberattacks, the tool's developer "Autumn-27" wrote Thursday on code-hosting platform GitHub.

"Given the misuse of the tool, the Artex project will no longer be updated and will be converted to closed-source," the developer said.

"No further versions will be released to the public, nor will maintenance support be provided."

Artex had been open-source -- allowing programmers to download its underlying code and customize it to suit their purposes.

Using Artex for cyberattacks was "entirely contrary to" the developer's intentions, they added, without referring directly to the alleged South Korean cases.

South Korea's Financial Services Commission says more than 68,000 people have been affected by the hacks.

Shinhan Bank, one of the breached institutions, said information attached to loan applications for about 25,000 customers had been leaked -- including names, phone numbers and annual income.

- 'Financially motivated' -

In Japan, meanwhile, around 20 companies have said their data may have been compromised in a spate of similar cyberattacks potentially impacting millions of customers.

The Japanese government has called on companies to strengthen their cyber defenses.

"At this stage, it is not clear what the background to these cases is or whether there are any links between them", Japan police chief Yoshinobu Kusunoki said on Thursday.

As AI makes hacking more sophisticated, "the scope of the damage is spreading across all areas on a scale that is difficult to compare with the past", South Korean President Lee Jae Myung said on Tuesday.

US cybersecurity giant CrowdStrike said on Wednesday that its investigations into digital clues suggested the attacker had used Artex, and was potentially a 26-year-old based in China.

"The threat actor is likely a Chinese speaker and financially motivated," a CrowdStrike blog post said.

Experts told AFP that Artex going closed-source would make the code harder for new users to obtain and adapt.

But the decision "cannot remove copies already downloaded or prevent people from continuing to use them", said Poe Zhao, founder of the analysis publication Hello China Tech.

"Because the code was public, people could download it, change it and run it themselves. The developer could ask users to follow the rules, but had little control over their actions," he added.

Ilya Kulyatin, CEO of Foundry Labs and founder of the Tokyo AI (TAI) tech community, said users of open-source programs can "remove restrictions built into the tool" which "makes certain forms of misuse easier".

"But openness also benefits defenders: researchers can inspect the code, identify weaknesses and improve protection."