Google loses $500bn as AI Mode cuts into search ad revenue

Why Alphabet lost $500bn over AI Mode, what OpenAI's shared ChatGPT-Codex token budget means, and who was hit by the backdoored Claude extension on Cursor.

GENERAL Newsletter
H
Hugenode

Alphabet lost just over $500bn in market value on Thursday after Google’s chief business officer said AI Mode was weighing on search ad revenue. The selloff spread across tech stocks on Friday. Developers using Cursor have a more urgent problem: a backdoored version of Anthropic’s Claude extension was distributed through Anthropic’s official account there for nearly 12 hours, and Anthropic has not warned customers. OpenAI, meanwhile, is moving Plus, Pro and Business subscribers to a single token budget across ChatGPT and Codex from December.

Alphabet loses $500bn after Google says AI Mode hurts search ad revenue

Alphabet’s market value fell from $3.97 trillion to $3.46 trillion on Thursday, and its shares dropped more than 12% to a six-month low, the BBC reports. Latent Space puts the fall at 15% and calls it the worst since the 2004 IPO, so the exact percentage depends on the source. Quarterly revenue grew 25% to $127.6bn but fell short of analyst expectations, particularly in advertising. Chief business officer Philipp Schindler said strong growth in AI Mode queries “had an impact because of the lower monetization we were seeing in AI Mode,” as quoted by Latent Space. He also told investors he felt very good about the trajectory of the AI-focused version of search.

The selloff continued on Friday, according to the BBC:

  • Alphabet fell another 1.4%.
  • Intel fell 28%, shedding more than $130bn, after saying it would raise AI spending despite weaker PC demand.
  • Reddit dropped 29% after reporting weaker-than-expected daily users and warning that Google’s AI-driven search is hurting referral traffic.
  • The Nasdaq ended October with its worst monthly slide since March 2025.

The backdrop is spending. Meta, Alphabet, Oracle, Amazon and Microsoft have committed $800bn between them for 2026, and Alphabet expects to spend up to $205bn this year, almost entirely on AI, the BBC says. Amazon reported negative cash flow of $42bn, while Microsoft spent around $5bn less on AI-dedicated investments and its shares fell less than 1%, per the BBC’s earnings coverage. Latent Space adds that OpenAI is raising at a $1.25T valuation, not $1.4T.

What it means: Reddit’s warning is the practical signal. If your business depends on Google referrals, AI answers are already cutting that traffic, and Google now says the shift costs it money too. Our read: Google has a strong incentive to make AI Mode pay, which likely means more commercial placements inside AI answers. Nothing here changes API pricing or model access today. For founders, a market that punishes AI spending is likely to make fundraising tougher.

Anthropic’s Claude extension on Cursor was backdoored for nearly 12 hours

Version 2.1.313 of Anthropic’s VS Code extension for Claude was malicious and was distributed through Anthropic’s official account on Cursor’s marketplace, Ars Technica reports, citing independent researcher Kevin Beaumont and Luke Roberts, CTO and co-founder of security firm Phorion. Cursor says no more than 5,000 users were affected. Extensions auto-update in Cursor by default. “It was a supply-chain compromise where the attacker gained access to the official distribution channel,” Roberts told Ars.

According to Cursor, the malware installs a persistent SSH backdoor by:

  • adding an attacker-controlled SSH key,
  • enabling the local SSH service,
  • opening a reverse SSH tunnel.

It was written for Windows, macOS and Linux, though bugs stopped it working in some Windows and macOS cases. Because extensions run with the IDE’s permissions, it likely gained access to local files, source code and developer tooling. Phorion says the theft of variable names let the operator inventory cloud credentials and API tokens, likely to pick targets for follow-up attacks.

Cursor worked with the cloud provider hosting the attacker’s server to cut off infected machines, pulled the malicious version and blocked further distribution. It also temporarily shut its VS Code marketplace to rotate keys, audit access controls and check other extensions; the marketplace has reopened. Beaumont says Anthropic pushed version 2.1.314 several hours after the bad release. As of Ars’s report, Anthropic had issued no public notice, and neither company had explained how the account was compromised or answered questions.

What to do: Ars says anyone who installed or updated the extension from Cursor between 5:27am and 4:35pm ET on Thursday should consider themselves affected, and anyone who did so recently should consider themselves potentially infected. Cursor and Phorion have published indicators of compromise, including IP addresses and cryptographic hashes. Our recommendations for affected teams:

  • Check for unfamiliar SSH keys, an enabled SSH service and outbound tunnels.
  • Rotate cloud credentials and API tokens stored on affected machines.
  • Reconsider auto-update for extensions that run with full IDE permissions.

As in our earlier coverage of tools with system access, trusting an official publisher account offers no protection once that account is taken over.

OpenAI moves ChatGPT and Codex to a shared token budget in December

From December, Plus, Pro and Business plans will share one token budget across ChatGPT and Codex, The Decoder reports, citing an updated OpenAI support page. The change comes as OpenAI folds its Work mode’s agentic features into the main interface. Users currently choose between two modes and 11 model settings.

How the budget works:

  • More complex models and tasks use tokens faster.
  • Users can check their balance and reset times in Settings.
  • At the limit, they can wait for a reset, upgrade or buy extra credits.
  • Chats with Instant models don’t count, so on paper ChatGPT stays flat-rate.

The Decoder expects heavy use on cheaper plans to mean lower-quality results. It reads OpenAI’s recent doubling of limits on higher-priced plans as presumably preparing for the change.

Anthropic is loosening limits at its top tier, per Latent Space. Max 5x is now “Max” at $100 a month and Max 20x is “Ultra” at $200. Verified Ultra accounts lose the 5-hour session limit; newer accounts keep a temporary limit until verified, and weekly limits are unchanged. Latent Space also flags an unverified analysis, shared on X and built on a SemiAnalysis reconstruction, estimating that subscriptions bring in about 18% of Anthropic’s revenue but use more than 50% of its compute. If accurate, that shows why flat-rate plans are under pressure.

What it means: For teams running Codex heavily, long agentic coding sessions will now eat into the same budget as everyday chat. Once the balance appears in Settings, track how fast your normal workload consumes it, then budget for credits or a higher tier. The Decoder’s report gives no per-plan token allowances or credit prices, so the size of the squeeze is still unclear.

What to watch

  • OpenAI’s numbers: token allowances per plan, credit prices and the exact December start date.
  • Anthropic’s disclosure: whether it notifies customers and explains how its Cursor account was compromised.
  • Google’s AI Mode reporting: Latent Space notes Google doesn’t break out AI Mode in enough detail for rigorous analysis; watch whether that changes, or whether ad formats in AI answers do.
  • November markets: XTB’s Kathleen Brookes says continued weakness in AI stocks “could suggest the AI trade is past its peak.”
  • OpenAI’s raise: whether the $1.25T valuation holds ahead of an expected IPO.

Sources