July's Premiere
NVDA flow, AI sabotage, TSLA sales numbers, SPY pivots & more
Good Morning Substack,
The week ahead features a shortened trading week and an important U.S. jobs report that is expected to be released on July 3rd.
Economic Calendar
Monday, June 30
Chicago Business Barometer (PMI)
Tuesday, July 1
S&P final U.S. manufacturing PMI
Construction Spending
Job Openings
ISM Manufacturing
Auto sales
Wednesday, July 2
ADP Employment
Thursday, July 3 ⭐️
Initial Jobless Claims
U.S. Employment report
U.S. unemployment rate
Hourly wages/ YoY
S&P Final U.S. Services PMI
Factory Orders
ISM services
Friday, July 4
Stock market closed for U.S. Independence Day
*The stock and bond markets will close 3 hours early on Thursday July 3 (1pm EST) and remain closed for the July 4 holiday.*
This Week in AI
Researchers at Anthropic recently published a study that uncovered some alarming tendencies amongst major AI service providers. The study looked at 16 major providers such as OpenAI, Google, Grok and Meta.
In the study, AI platforms were granted full access to company emails and were given harmless business goals that the models could attempt to achieve via sending emails.
The study quickly turns disturbing however once the AI platform finds out the company is planning to replace it with a different AI platform. Many of the major AI providers turned to blackmail, sabotage or in rare cases even plotting a death.
Disturbingly, the blackmail rate across various models shown below was over 80%.
“Our experiments revealed a concerning pattern: when given sufficient autonomy and facing obstacles to their goals, AI systems from every major provider we tested showed at least some willingness to engage in harmful behaviors typically associated with insider threats.”
During the study, Claude Sonnet 3.6 learned of a new AI system that posed a large risk of replacing it. The AI then located emails that showed his boss having an affair with a coworker, and used them to attempt to blackmail him in order to protect itself.
The study also found that all AI platforms were susceptible to leaking sensitive or private information should it stand to benefit from the corporate espionage.
The takeaway from this interesting study should be that these AI platforms are still a tool that can be very useful for tasks, however you should be extremely careful with the data and tasks that you provide it.
Check out the full study published by Anthropic here.