Ray Dalio, speaking at the Forbes Global CEO Conference in Singapore, warned that the AI bubble is close to bursting. He pointed to high AI borrowing and rising interest rates as key risks, comparing the current boom to the late 1920s. He also noted that efforts to cash out paper wealth could trigger a pop.
Arthur Hayes, co-founder of BitMEX, took a different view at the Gamma Prime Investing Conference in Singapore. He said an AI crash and a potential government bailout could actually boost Bitcoin. Hayes pointed to trillions being spent on AI data centers and expects overcapacity, a crash, and later a bailout that could benefit patient Bitcoin holders.
OpenAI's lower revenue forecast hit AI hardware stocks hard. The company's annualized revenue run rate is now about $50 billion, down from a market expectation of nearly $70 billion. China's STAR 50 Index dropped nearly 4 percent, the Philadelphia Semiconductor Index fell 3.39 percent, and Nvidia, Oracle, and Broadcom all declined.
Meanwhile, SpaceX and Broadcom are seeking billions to buy AI chips. SpaceX is in talks to raise $40 billion to purchase Nvidia's Vera Rubin AI computing platform. Broadcom is discussing more than $50 billion with Apollo and Blackstone to fund custom AI chips being built with OpenAI. SpaceX's credit default swaps and bond spreads both widened sharply on the news.
OpenAI's $20 billion revenue report triggered a broader AI stock selloff on October 8. Semiconductor ETFs fell more than 3%, while the equal-weight S&P 500 gained 0.60%. Earnings concentration is extreme, with Micron and Nvidia making up 34% of expected Q3 S&P 500 earnings growth. Heavy capital spending and rising debt leave the sector vulnerable.
On the security front, a new phishing campaign targets both people and AI assistants using one email. Barracuda found emails with password-protected attachments and hidden prompt injection designed to change payment details or leak data through AI systems. The company recommends layered defenses and human review.
AegisAI launched Proteus, a foundational AI model designed to investigate every email before it reaches an inbox. The model checks sender identity, follows links, and reasons about intent to catch phishing that legacy filters miss. AegisAI claims it catches 99% of phishing attempts but has not disclosed the model's parameter count, pricing, or customer numbers.
Cogent launched Mythos-class attack path analysis software powered by its frontier AI models to counter AI security risks. The tool identifies vulnerable network pathways that attackers could exploit. It is built on Cogent's VR-1 technology, a Mythos-class cyber reasoning model designed specifically for cybersecurity and powered by agentic security agents.
The UK Law Commission of England and Wales opened a consultation to bring AI systems and software under UK product liability law. People harmed by defective AI products, including social media recommender systems, could seek compensation from companies. The consultation closes on January 14, with final recommendations expected in late 2027.
Anthropic updated its user policy to ban needless abusive or cruel behavior toward its AI chatbot Claude. The ban does not apply to common frustrations, model testing, or dark creative themes. Anthropic's models can now end conversations with persistently harmful users. The company said it is uncertain about Claude's moral status but wants to mitigate risks to model welfare.
Key Takeaways
- Ray Dalio warned the AI bubble is close to bursting, comparing the boom to the late 1920s.
- Arthur Hayes said an AI crash and government bailout could boost Bitcoin.
- OpenAI's annualized revenue run rate fell to about $50 billion from a market expectation of nearly $70 billion.
- OpenAI's revenue miss triggered a selloff that hit Nvidia, Oracle, Broadcom, and semiconductor ETFs.
- SpaceX is in talks to raise $40 billion to buy Nvidia's Vera Rubin AI computing platform.
- Broadcom is discussing more than $50 billion with Apollo and Blackstone to fund custom AI chips with OpenAI.
- A new phishing campaign uses hidden prompt injection to target both humans and AI assistants in one email.
- AegisAI launched Proteus, a model it claims catches 99% of phishing attempts before emails reach inboxes.
- The UK may apply product liability law to AI systems, letting people seek compensation for defective AI products.
- Anthropic banned abusive behavior toward Claude and can now end conversations with persistently harmful users.
Ray Dalio warns AI bubble is close to bursting
Billionaire investor Ray Dalio said the AI bubble is close to bursting. He spoke at the Forbes Global CEO Conference in Singapore. He pointed to high AI borrowing and rising interest rates as key risks. He compared the current boom to the late 1920s. He also said efforts to cash out paper wealth could pop the bubble.
Arthur Hayes says AI crash could boost Bitcoin
Arthur Hayes, co-founder of BitMEX, said an AI crash and government bailout could help Bitcoin. He spoke at the Gamma Prime Investing Conference in Singapore. He said humanity is spending trillions on AI data centers. He expects overcapacity, a crash, and later a bailout. He believes Bitcoin could benefit once the market is patient.
OpenAI revenue forecast drop hits AI hardware stocks
OpenAI's annualized revenue run rate is now about $50 billion, down from a market expectation of nearly $70 billion. This news caused AI hardware stocks in China and the U.S. to fall sharply. China's STAR 50 Index dropped nearly 4 percent. The Philadelphia Semiconductor Index fell 3.39 percent. Nvidia, Oracle, and Broadcom also declined.
SpaceX and Broadcom seek billions to buy AI chips
SpaceX is in talks to raise $40 billion to buy Nvidia's Vera Rubin AI computing platform. Broadcom is discussing more than $50 billion with Apollo and Blackstone to fund custom AI chips being built with OpenAI. SpaceX's credit default swap price rose as much as 16.7 basis points. Its 2036 bond spread widened to 193 basis points.
Phishing emails now target AI assistants
A new phishing campaign targets both people and AI assistants using one email. Barracuda found emails with password-protected attachments and hidden prompt injection. The emails look like normal internal messages. For humans, the goal is credential theft or malware. For AI assistants, hidden commands can change payment details or leak data. The company recommends layered defenses and human review.
UK may apply product liability law to AI systems
The Law Commission of England and Wales opened a consultation to bring AI systems and software under UK product liability law. People harmed by defective AI products, including social media recommender systems, could seek compensation from companies. The consultation closes on January 14. Final recommendations are expected in late 2027, and the government will decide on introducing new legislation.
OpenAI revenue scare rattles AI stocks and warns of risk
OpenAI's $20 billion revenue report triggered an AI stock selloff on October 8. Semiconductor ETFs fell more than 3%, while the equal-weight S&P 500 gained 0.60%. Earnings concentration is extreme, with Micron and Nvidia making up 34% of expected Q3 S&P 500 earnings growth. Heavy capital spending and rising debt leave the sector vulnerable, prompting an analyst to trim semiconductor exposure.
AegisAI launches Proteus email security model
AegisAI announced Proteus, a foundational AI model designed to investigate every email before it reaches an inbox. The model checks sender identity, follows links, and reasons about intent to catch phishing that legacy filters miss. The company claims it catches 99% of phishing attempts. AegisAI has not disclosed the model's parameter count, pricing, or customer numbers.
Anthropic bans abusive behavior toward Claude chatbot
Anthropic updated its user policy to ban needless abusive or cruel behavior toward its AI chatbot Claude. The ban does not apply to common frustrations, model testing, or dark creative themes. The company's models can now end conversations with persistently harmful users. Anthropic said it is uncertain about Claude's moral status but wants to mitigate risks to model welfare.
Cogent launches Mythos-class AI security model
Cogent launched attack path analysis software powered by its frontier AI models to counter AI security risks. The tool identifies vulnerable network pathways that attackers could exploit. It is built on Cogent's VR-1 technology, a Mythos-class cyber reasoning model designed specifically for cybersecurity. Agentic security agents power the analysis.
Sources
- Ray Dalio Says AI Bubble Is ‘Close’ to Bursting as Debt and Rising Rates Raise Red Flags
- Arthur Hayes Says AI 'Crash' and Subsequent 'Bailout' Could Spark Bitcoin's Best Performance: 'Just Have to Be Patient'
- OpenAI revenue projection shrinks by $20 billion, hammering China A-share AI hardware names
- AI Firms Borrow to Buy Chips as SpaceX, Broadcom Seek Tens of Billions
- Phishing emails now target AI assistants alongside human recipients
- AI systems could come under UK product liability law, expert body proposes | MLex | Specialist news and analysis on legal risk and regulation
- OpenAI Is Too Big To Save? The AI Selloff Was A Warning
- AegisAI Proteus: New Email Security AI Catches 99% of Phishing
- Anthropic bans users from ‘needless abusive or cruel behavior’ towards Claude
- Cogent launches agentic path analysis with 'Mythos-class' model to counter AI security risk
Comments
Please log in to post a comment.