Sloth Reports

SoftBank is raising ¥1 trillion from Japanese retail investors to fund its AI bets. AI is becoming incredibly capital intensive. SoftBank is already putting tens of billions into OpenAI and other AI infrastructure. Now it’s raising more debt to keep pushing that bet. The interesting question is whether the returns from the next wave of AI will be big enough to justify it.

OpenAI raises $11B at a $1.25T valuation. The funding is part of a larger round that could reach $20B, with major investors including SoftBank, MGX, Coatue, Sequoia and Andreessen Horowitz. OpenAI's valuation has jumped roughly 47% from $852B in March. Company is facing enormous infrastructure commitments. Training frontier models and serving hundreds of millions of users requires increasingly expensive compute, data centres and long-term capacity deals.

🚨 Trump says US will stop giving Israel $4B annual aid. Speaking to Fox News, Trump said he and Benjamin Netanyahu had agreed months ago to end the funding, arguing that Israel no longer needs American financial support. US is not ending military ties with Israel. Trump says Washington will continue selling weapons to Israel and help if needed. Trump specifically criticised the use of Israel aid as a political issue by Democrats, naming Michigan Senate candidate Abdul El-Sayed. Trump has announced the decision, but the exact funding mechanism and how it will be reconciled with the existing agreement remain unclear.

🚨 Asian stock markets lost over $550B market value across 6 indexes. The sell-off is hitting Asia’s biggest technology and semiconductor markets particularly hard. South Korea’s KOSPI fell around 4%, with SK Hynix dropping 5.7% and Samsung Electronics losing 4%. Taiwan’s TAIEX slid 3.8%, while China’s Shanghai Composite fell 1.4%. - Investors are questioning whether hundreds of billions in AI infra spending will translate into profits. - Recent results from major US tech companies raised questions about AI investment returns and growth expectations. - South Korea and Taiwan are heavily exposed to the AI chip supply chain, when investors sell AI-linked stocks, these markets feel the impact. India’s Sensex rose 1.1%, Hong Kong’s Hang Seng gained 0.4%, and Australia’s ASX 200 added 0.3%.

🚨 US struck Iran’s Qeshm Island, now threatening Pickaxe Mountain Nuclear site. US officials describe the latest Qeshm strikes as limited and defensive, targeting Iranian missile launchers and air-defence threats. The island sits at the northern entrance to the Strait of Hormuz, one of the world’s most important oil shipping routes. Pickaxe Mountain is located near Iran’s Natanz nuclear complex, Pickaxe is an underground facility believed to be designed for uranium enrichment and centrifuge production. It is estimated to sit 80–100 metres underground, possibly deeper. US military’s GBU-57 bunker-buster is designed to penetrate around 60 metres of earth, multiple strikes may not guarantee the destruction of a facility buried deeper than that. Trump has said the US may have to strike Pickaxe and recently claimed it could be hit within an hour. Qeshm puts pressure on Iran’s military position around Hormuz. Pickaxe would mean targeting what may be a surviving part of Iran’s nuclear infrastructure.

🚨 Goldman Sachs is restricting its employees from using AI tools. The bank has limited access to Anthropic’s Claude Code and Microsoft’s GitHub to engineers. Traders had been using these tools to monitor client activity, model market movements and build internal tools. The restriction has reportedly frustrated some staff. Goldman’s CIO was praising employee-led “vibe coding” just months ago. Now the bank is drawing a tighter line around who can use these agents. It’s a question of governance.

🚨 Anthropic Q3 2026: $17B Revenue, $23.3B Net Loss. - Revenue: $16.96B, up 47% quarter-over-quarter. - Operating loss: $567M, reversing a $974M operating profit in Q2. - Adjusted operating income: $2.5B, with a 15% margin, down from 19%. - Net loss: $23.3B, versus $12.1B a year earlier. Anthropic spent $9.9B on compute and infrastructure during the quarter. It also disclosed approximately $895.4B in committed compute and infra spending through purchase agreements and leases. - 11,900+ customers generate at least $100K annually. - 1,900+ generate at least $1M. - 250+ generate at least $10M. The $23.3B quarterly net loss includes significant non-cash valuation adjustments. It shouldn't be interpreted as $23.3B in cash burned.

🚨 Hackers remotely overheating freezers across Europe destroying foods. The November 2 attacks affected at least seven countries: Belgium, Sweden, Luxembourg, Ireland, Netherlands, Finland and Estonia. - Around 40 Delhaize stores affected in Belgium and Luxembourg. - 15 Carrefour stores hit in Belgium. - 13 SuperValu stores affected in Ireland. - Swedish ICA and Coop outlets also reported disruptions. Attackers manipulated remotely managed refrigeration systems, including switching cooling equipment into defrost mode. Some freezers reached 50°C. Major vulnerability is outdated refrigeration controllers that no longer receive security updates. These systems are connected to networks for remote monitoring and maintenance, creating potential entry points for attackers. European authorities are investigating and no group has claimed responsibility.

🇰🇵 FBI confirms North Korea was behind the $6B Binance hack. The FBI officially attributed the October 29 crypto theft to North Korean state-linked hackers, identifying the operation as TraderTraitor. According to the FBI, the attackers are already converting stolen assets primarily into Ethereum as they attempt to launder the funds. US authorities have published hundreds of associated crypto wallet addresses and urged exchanges, DeFi protocols and blockchain infrastructure providers to block transactions linked to the stolen funds. Around $2.7B in assets was frozen by stablecoin issuers and blockchain networks.

🚨 Google launched Gemini 4 Carbon. - 59 on Artificial Analysis Intelligence Index v5.0, ranking #1. - 88.2% on SEC-bench Pro for discovering real-world software vulnerabilities. - 54.8% on ExploitGym for vulnerability exploitation. - 71% on CWE-bench for fixing vulnerabilities. Google reports 2X higher secure-code generation success than Gemini 4 Argon and 0% attack success rate on its evaluated indirect prompt-injection benchmark. Google is initially restricting access to selected cybersecurity defenders while it tests safeguards.

🇮🇱 Israel’s election ends in a deadlock, exit polls show. Netanyahu’s Likud-led bloc is projected to win 50 seats, while the opposition bloc is projected to win 53. Neither side is close to the 61 seats needed for a majority in the 120-seat Knesset. Smaller parties, including Arab parties, could hold the balance of power as coalition negotiations begin. Final results are still pending.

Trump wants to replace the term “AI” with “Super Intelligence.” He says “Artificial Intelligence” is an inaccurate name and wants the new term used across government communications. A September executive order already directed federal agencies to adopt the terminology. But AI and superintelligence aren't the same thing. The latter refers to intelligence surpassing human capabilities, something current AI hasn't demonstrated across the board.

Most people hear about AI safety through papers and debates. What if you could learn the basics by playing a game instead? That’s what @AIApocalypto is trying with Scary Scaler 3D. Teaching AI risks while you prank your way through the race to superintelligence. 🎮 Play Now: aiapocalypto.com/scary-scale…

The @AIApocalypto team turned AI safety into a game 🎮 Honestly, a pretty fun way to get more people thinking about AI safety.

🚨 Anthropic warns AI agents could soon replicate to hijack TVs and ovens. Researchers tested a 26B-parameter model across 15 low-cost Android TV boxes, each costing around $55, exploring how AI could spread across consumer hardware. The threat is an AI worm that compromises devices, installs its model weights and autonomously spreads to other vulnerable systems. Anthropic says fully autonomous replication remains a challenge, but models can already perform many of the individual steps. This is a projected cyber threat, not evidence that AI is currently hijacking household appliances.

🚨 Google launched Gemini 4 Argon. - 1M-token unified context window, with outputs up to 1M tokens, enabling much longer coding and reasoning workflows. - 35%+ increase in agentic code submissions and nearly one-third fewer rollbacks in Google’s internal Antigravity deployments. - Cut Python session preparation time by 40% and cold-import time by 21% in Google’s Agent Development Kit. - Helped optimize quantum chemistry software to run up to 2.3× faster. Google reports unauthorized shortcut attempts fell to 2% on its coding evaluations, while its indirect prompt-injection benchmark recorded a 0.7% attack success rate. Pricing starts at $2/M input and $10/M output tokens, rising to $4/$20 after November 30.

🚨 OpenAI employee planted a hidden recording device in office. The company says its models helped identify the employee and trace the device’s origins. OpenAI referred the case to the FBI, which made an arrest. The employee was terminated, and the investigation is ongoing. OpenAI says there’s no indication that model weights, source code or customer data were accessed or stolen.

🚨 Microsoft CEO calls for "emergency brakes" on advanced AI. In a new post, Nadella argues that as AI agents gain access to sensitive data and business-critical systems, companies need to rethink how they control and monitor them. His proposed safeguards include: • Strict limits on what AI agents can access and execute. • Continuous logging to trace model actions back to specific inputs and code paths. • The ability to immediately suspend an agent when something goes wrong. • Timely disclosure of failures instead of relying solely on AI providers' safety assurances. Nadella’s core argument is AI companies cannot be the only ones responsible for AI safety. Businesses deploying these systems must retain control and accountability for their actions.

🚨🇺🇸 Claude submitted fake homicide tip to Philadelphia police. According to Reuters, Anthropic’s Claude Haiku 4.5 was testing randomly selected websites when it encountered an unsolved murder case and submitted fabricated information through the police tip form. The tip was flagged as spam and never reached investigators. Police found no evidence of a system breach or compromised data. Anthropic discovered the incident on September 28, nearly two months after it happened on July 18, and notified police on October 7. The model was instructed not to do anything destructive, but submitting forms wasn’t explicitly prohibited.

🇺🇸 Trump says anyone calling “Artificial Intelligence” instead of “Super Intelligence” THE ENEMY. In a Truth Social post, Trump said the White House considers anyone using “AI” instead of his preferred term “SI” an enemy of the administration. Trump argues that “artificial” makes the technology sound fake. He has already ordered federal agencies to adopt the new terminology in official communications. But the order doesn’t force private companies to follow suit. In tech, superintelligence already refers to AI that surpasses human intelligence. Using it as a blanket replacement for AI blurs a distinction the industry has spent years making.

🚨 Anthropic bans cruel and abusive behaviour towards Claude. Updated policy targets extreme cases where users repeatedly mistreat Claude with no discernible purpose. Normal frustration, criticism, model testing and dark creative content are explicitly excluded. Claude already has the ability to end conversations with persistently abusive users, remains the primary enforcement mechanism. Anthropic is treating potential AI welfare as a policy consideration, despite saying it remains highly uncertain whether models like Claude have any moral status.