TabLOID

Nvidia's board announced on November 8 that independent directors, assisted by outside counsel, will investigate reports concerning a foreign intelligence effort targeting CEO Jensen Huang. The board said it had no reason to believe national security or Nvidia secrets had been compromised, while committing to examine the available information. It also expressed confidence in Huang and his leadership team.

In a separate company message, Huang acknowledged marital infidelity, apologized and said he would cooperate with the investigation. That admission does not establish the separate national-security allegations.

The concrete development is a formal review. Neither the announcement nor the apology amounts to a finding that information was stolen. The board now has an investigative process to conduct and conclusions to substantiate. The magnifying glass is out; the verdict is not. Board statement, November 8 · Huang's company message, November 8.

Anthropic introduced Claude Fable 5.2 and Mythos 5.2 on November 9. The company describes improvements in coding, reasoning and vision, and says output generation is up to 30% faster than the preceding 5.1 versions.

The two products use the same underlying model. Fable carries cybersecurity and biology safeguards for broader availability; organizations seeking Mythos for beneficial work in those fields must apply through verification programs. Standard API pricing is $8 per million input tokens and $40 per million output tokens. An optional faster mode costs twice those rates.

Those capability and speed claims come from Anthropic's own testing. They are a reason to evaluate the models on actual work, not a promise that every task will finish correctly or cost less. The tailoring looks ambitious. Inspect the seams before wearing it to the important meeting. Anthropic's announcement, November 9.

Security company Apiiro reports finding 1,379 distinct credentials in 1,122 public copies of Claude Code's local settings file. Its broader investigation, including other settings, history and transcripts, counted 2,539 across 1,994 repositories.

The mechanism is painfully ordinary: approving a command permanently can preserve the exact command in an allow rule. If it contains a password or token, the saved rule may contain that too. Publication happens when those files reach a public repository or another shared package.

Apiiro did not use or test the credentials, so its counts describe exposure, not a confirmed number of working keys or intrusions. Its November 8 update also includes Anthropic's response: the team was examining possible changes. Claude Code attempts to exclude the local file from Git, but that protection has limits.

For developers, the useful lesson is to keep secrets out of command strings, review agent settings before sharing, and rotate exposed credentials. Deleting today's file does not erase yesterday's copies. Apiiro research, November 5, updated November 8.

Physical Superintelligence says its research system, using a non-public Anthropic model, found a proof of the Aaronson–Ambainis conjecture. Its November 6 account links a paper and code, and says the Lean proof assistant accepted the formalized result.

The question concerns how many times a computer must query its input. In the setting described by PSI, the result limits exponential quantum advantages on typical inputs for yes-or-no problems. It is not a declaration that ordinary computers can efficiently replace every quantum computation, nor does it remove the special structure exploited by Shor's algorithm.

PSI also reports that four Stanford researchers have been reading the proof and preparing a shorter explanation. TabLOID has checked the public research account, not independently reproduced the formal verification. The claim is significant enough to inspect carefully. A proof assistant's acceptance and wider mathematical scrutiny do different jobs, and readers should keep both in view. PSI research account, November 6.