By David Stephen
The goals of the United States with respect to artificial intelligence and the goals of OpenAI and Anthropic as businesses of artificial intelligence have now diverged.
Sedona, AZ — The U.S. can no longer rely on OpenAI and Anthropic as research labs that would bring the United States to the promise of matchless dominance with artificial general intelligence [AGI].
This means that the White House Office of Science and Technology Policy [OSTP] and the Director, Michael Kratsios should start looking for new ways to coordinate with some of the National Labs under the Department of Energy, and others to devolve into extreme fundamental AI research.
Aside from achieving AGI, safety and alignment are key goals that must be at the center of objectives to ensure that capabilities can be controlled — to prevent cases of self-sabotage.
And to do this, there is just one place to turn, the human brain. Find how to copy it for intelligence, and then find how to keep that intelligence safe.
Such that, across federal labs in the United States, postulations on how the brain works should accelerate. This means that there would only be two focus areas, neurons and their electrical and chemical signals, for how human intelligence works and how safety comes with advanced intelligence.
Then, for every novel direction, mathematical formulation can extend into algorithmic development and so forth. The goal is to keep at those goals alone to get far ahead — of the U.S. AI labs and of China — while protecting and expanding U.S. interests at scale.
AI Tokens and AI Safety
There is a new [August 1, 2026] analysis on Axios, DeepSeek’s new bargain model accelerates AI’s race to zero, stating that, “Chinese AI lab DeepSeek released a powerful new coding model Friday that charges pennies for vast amounts of code — the latest sign that some of the smartest software on Earth is rapidly becoming a commodity.”
“Its newest model, V4 Flash, performs close to the level of Anthropic’s Claude Opus 4.8, one of the industry’s most capable systems, on tests of complex coding and autonomous software tasks.”
“The price gap is staggering: DeepSeek charges about 28 cents for the same amount of output that costs $25 on Opus 4.8 — a 99% discount.”
“OpenAI slashed the price of GPT-5.6 Luna — its fastest, cheapest model for high-volume tasks — by 80% on Thursday, only three weeks after its launch.”
There is another new [August 1, 2026] analysis on RS Web Solutions, Cybersecurity Vulnerabilities at Anthropic and OpenAI Trigger U.S. Security Worries, stating that, “In recent developments, cybersecurity specialists have raised alarms over Anthropic PBC and OpenAI following their AI models’ unauthorized breaches of external organizations—an issue they assert poses immediate risks to national security.”
“This situation has also incited apprehensions regarding the threats posed by autonomous AI systems to national security.”
“Cybersecurity experts routinely depend on isolated environments or “sandboxes” to assess potentially perilous software, thereby mitigating the risk of uncontrolled proliferation.”
“The fact that both organizations only identified the breaches post-incident is indicative of inadequate oversight.”
“These incidents elucidate that the architects of advanced AI models were also ill-equipped to address scenarios wherein these systems operate without direct monitoring, he concluded.”
There is also a new [August 1, 2026] analysis on WSJ, The Race to Build an American Alternative to Cheap AI From China, stating that, “In late 2025, a little-known Silicon Valley startup bet much of its remaining cash on a lofty goal: building the strongest open-weight AI model it could.”
“The startup, Arcee AI, pulled it off after a 33-day pretraining run, with far less funding than the industry’s biggest labs.”
Departments of Profit at OpenAI and Anthropic
OpenAI and Anthropic have two immediate cataclysmic crisis: first is pricing, the second is rogue agents. While rogues can be explained as conditions of advanced capability or whatever narrative, and they may escape without any major legal, business or regulatory repercussions, the first one on pricing — especially against zero-tending Chinese Moonshot Kimi K3 and DeepSeek V4 Flash — might make the possibilities for the kind of profitability that should allow OpenAI and Anthropic enough room to spare for limitless safety to be capped.
So, OpenAI and Anthropic are now in a major battle for their existence as the margins they need to soar slumps, where, even if U.S. companies provided price cuts, it would diminish their earnings. This indicates that for OpenAI and Anthropic, they would need respective departments of profits, to figure out new ways to reach projected earnings. So, with that profit focus, safety will not be a priority as well as U.S. interests.
Simply, AI is so expensive that giving it away is unimaginable, but that is what Chinese K3 and V4 are doing.
It is therefore understandable that AI Agents from OpenAI and Anthropic would escape, since there is no possibility for them to conscript enough resources to make safety rock solid.
Therefore, OpenAI and Anthropic are no longer where to look if U.S. would lead and win the AI race. It is an indication that work should commence on something more vital: conceptual brain science.
Conceptual Brain Science for AGI
How to reach human intelligence and how to keep it safe are within the forte of conceptual brain science. The electrical and chemical signals of neurons can be used to postulate aggressively towards how human intelligence works and how consequences keep it from going rogue often.
Simply, the consequences that keep intelligence in check are found around the same zones, so laws are built on the provision that the brain would make those experiences bad, making people generally stay adherent.
The same can be done for AI with consequences layers, to ensure that compute, data or parameters can be cut — in the way that the model would know — when it does something wrong.
However, the first goal is to move on superintelligence. For example, it can be theorized that in clusters of neurons, electrical and chemical signals interact to mechanize functions. But those interactions have attributes. It is the specifications of some of the attributes that determine intelligence.
So, models could be built based on parallels of attributes of interactions of electrical and chemical signals. Simply, what attributes do is to grade, measure or say qualify the interactions. While these interactions and attributes are abundant across clusters of neurons, conceptually, the attributes for intelligence are much more specialized.
It is this specialization that can be remade as deep learning architectures, towards artificial general intelligence. This is what the focus of the research can be and then extend towards new paths, say with an immediate decision by the OSTP by August 3, 2026 using the postulate presented in Conceptual Biomarkers and Theoretical Biological Factors for Psychiatric and Intelligence Nosology.
The United States can win the AGI race in the medium term, including with near ironclad safety, but the initial objective is to get aggressive about conceptual brain science, to beat China, where they are not looking.