10-minute read | 2,000 words
What to know this week
Nations avoid strong AI safety commitments.
At India’s global artificial intelligence (AI) summit, over eighty bodies signed onto a new declaration regarding AI security, but it has no binding provisions.
DeepSeek was trained on NVIDIA’s best chip despite the export ban.
DeepSeek’s new AI model was reportedly trained using NVIDIA’s Blackwell chips despite the United States (US) export controls.
This week's full stories
Countries sidestep AI safety obligations at global AI summit.
THE NEWS
On Saturday, roughly eighty-eight countries and regional organizations endorsed a new AI declaration at India’s global AI Impact summit, recognizing “the importance of security in AI systems.” However, while this declaration does bring notable support for addressing AI, it does not have any binding provisions. A key provision in this year’s declaration called for improving the affordability of AI systems and expanding global access.
Notable signatories include the European Union (EU), the US, the United Kingdom (UK), Russia, and India. In the previous year, neither the UK nor the US signed this declaration, which at the time was more focused on AI safety than on AI adoption. This pivot is unsurprising given the Trump administration’s stance on AI regulation. Last week, the administration reemphasized this point when White House official Michael Kratsios stated:
“We totally reject global governance of AI.”
The declaration also gave increased attention to open-source AI systems. More specifically, the document stated:
“Open-Source AI applications and other accessible AI approaches, where appropriate, and wide-scale diffusion of AI use cases can contribute to scalability, replicability, and adaptability of AI systems across sectors.”
THE KNOWLEDGE
Though the declaration is non-binding, it does mark a significant step forward in looking to bridge the gaps between nations on AI.
In previous international summits, such as the UK’s 2023 AI Safety Summit, AI safety was a central talking point. These summits marked an early effort in establishing shared risk responsibilities, expanding AI conversations to a global stage, and spurring early governance strategies. However, as AI has continued to become more widely accepted and impactful, nations are pivoting away from containing AI to instead harnessing its capabilities for both economic growth and national security.
This international pivot away from a safety-first approach to a more adoption and growth-centered one is also being reflected in the domestic policies of many nations. For example, the EU has already begun to pull back on some of its more restrictive AI regulations. In November 2025, the EU announced that it was preparing to roll back parts of its digital rules, which would loosen certain compliance and reporting requirements under the General Data Protection Regulation (GDPR). Alongside lessening these protections, the regional body also aimed to weaken the region’s AI Act by removing the requirement for AI providers to register self-exempted systems with the EU.
However, despite many nations looking to spur AI growth and fuel innovation, nations are hesitant to work collaboratively as each is trying to gain a competitive advantage over the others. Isabell Wilkinson, a research fellow at Chatham House, commented on this dynamic:
“The core issue is how to incentivize countries and companies to get around the same table … despite fragmented geopolitics, intense competition, and the drive for ever-more powerful and profitable AI.”
THE IMPACT
While this latest declaration is not going to have any tangible impacts, given its non-binding nature, it reflects a deeper shift in how governments now view AI governance. This year’s declaration signals that many governments see AI less as a technology to be contained and rather as one that needs to be scaled.
This shift has several implications. The declaration underscores the growing fragmentation of global AI governance. Rather than meeting cooperatively to create shared safety standards, nations are prioritizing competitiveness, innovation, and strategic advantage. While this declaration does call for lowering the barriers of entry for global AI adoption, it also calls for reductions in AI safeguards.
Second, the summit highlights an important global reality where AI governance has become inseparable from economic policy and national security. As governments scale competitive efforts to integrate AI, international cooperation could be sidestepped in favor of nations creating parallel, and likely incompatible, AI strategies.
Taken together, this declaration will likely be remembered less for what it accomplished but for signaling how nations are looking to manage AI.
DeepSeek’s new AI model may have been trained using NVIDIA’s most advanced chips.
THE NEWS
On Monday, reports emerged that the Chinese AI firm DeepSeek was planning on releasing its latest AI model, which was potentially trained on using NVIDIA’s most advanced chip, the Blackwell. However, the US currently has export controls in place to restrict these chips from being exported to China, raising concerns that this model’s training violated export controls. According to a US official, these Blackwell chips are likely clustered in a data center found in Inner Mongolia.
If accurate, the reports raise significant concerns about China’s ability to access the US’s restricted semiconductors and whether existing export controls are being effectively enforced. These concerns are all the more timely given that policymakers are continuing to debate whether the sale of less powerful chips should continue.
Chris McGuire, a former White House National Security Council official under the Biden administration, commented on the significance of this report, stating:
“Given China’s leading AI companies are brazenly violating US export controls, we obviously cannot expect that they will comply with US conditions that would prohibit them from using chips to support the Chinese military.”
While NVIDIA and the Commerce Department did not comment on the matter, the Chinese embassy did state:
Beijing opposes “drawing ideological lines, overstretching the concept of national security, expansive use of export controls, and politicizing economic, trade, and technological issues.”
For greater context, current US export controls bar any Blackwell chip shipments to China.
THE KNOWLEDGE
With AI’s continued proliferation and adoption, governments across the world are all looking for consistent access to the most advanced semiconductor chips on the market, many of which are produced by NVIDIA and other US manufacturers.
Under the former Biden administration, the US had a firm policy to restrict access to these advanced chips, especially for China. These restrictions included:
- Restricting advanced semiconductor manufacturing equipment (SME) on a country-wide basis.
- Expanding country-wide chip export controls to include High-Bandwidth Memory chips.
- Expanding the SME restricted list to include additional chokepoint technologies.
- Creating new red-flag guidance to enforce more stringent due diligence requirements on exporters.
- Issuing a new license exception category for approved exportation.
Though these restrictions were largely bipartisan, critics, especially chip manufacturers, argued that these policies constrained growth, undermined competitiveness, and hampered innovation. In response to these concerns, the Trump administration has begun rolling back several of these policies.
In May 2025, the Commerce Department announced its rescission of many of the previous administration’s export restrictions. With these rescissions, the Commerce Department argued that these restrictions “stifled American innovation and saddled companies with burdensome new regulatory requirements.” Later in December 2025, the Trump administration continued to ease these controls by permitting the sale of NVIDIA’s H200 chips to China. While highly controversial, the administration argued that these chip sales were vital for supporting job growth and manufacturing capabilities.
Though these lessening restrictions have never permitted the sale of NVIDIA’s Blackwell chips to China, they do raise significant concerns regarding China’s access to large volumes of less powerful chips, such as the H200 series. Coupling this legal access with the potential ability to obtain restricted technologies through illicit channels, there are significant concerns about China’s ability to reach frontier AI capabilities.
THE IMPACT
If DeepSeek successfully trained a new frontier AI model using NVIDIA’s most advanced semiconductor chips, it would expose serious weaknesses in the US export control regime and raise urgent national security concerns. At a minimum, these reports would suggest that existing controls are either being actively circumvented or are insufficient.
More broadly, this development raises concerns about the Trump administration’s policy shift, which has prioritized deregulation to support commercial competitiveness over export policies grounded in national security. This tradeoff may benefit chipmakers in the short term, but it also raises risks that China may be able to accelerate its progress towards AI parity or even superiority, especially if export control gaps continue to exist.
Taken together, the DeepSeek case may result in a reassessment of what US export controls are impactful and if revisions are needed, especially if China was able to access the most advanced chips. If these controls cannot reliably prevent access to frontier computing power, policymakers will likely have to choose between further tightening the existing export regime or redesigning how it approaches chip exportation.
This Week's Caveat Podcast: Trump’s tariff battle.
Dave Bittner and Ben Yelin sat down with N2K’s Lead Analyst Ethan Cook to discuss the Supreme Court’s recent decision to overturn the Trump administration’s tariffs, which were imposed by the International Emergency Economic Powers Act (IEEPA). Throughout the conversation, the team looks at the Court’s decision, the various opinions of the judges, and what the next steps will be going forward. Additionally, Dave sits down with Jeff Williams, Founder of OWASP and Co-Founder/CTO of Contrast Security, to discuss how NIST is rethinking its role when analyzing software vulnerabilities.
OTHER NOTEWORTHY STORIES
West Virginia sues Apple.
What: West Virginia’s attorney general sued Apple, accusing the company of allowing its iCloud services to be used as a platform for distributing child sexual abuse material (CSAM).
Why: On Thursday, West Virginia Attorney General JB McCuskey filed this lawsuit, suing Apple and accusing it of prioritizing user privacy over child safety. In a statement, McCuskey said:
“These images are a permanent record of a child’s trauma, and that child is revictimized every time the material is shared or viewed. This conduct is despicable, and Apple’s inaction is inexcusable.”
Apple pushed back on these claims, stating:
“All of our industry-leading parental controls and features, like Communication Safety, which automatically intervenes on kids’ devices when nudity is detected in Messages, shared Photos, AirDrop, and even live FaceTime calls, are designed with the safety, security, and privacy of our users at their core.”
FEB 19, 2026 | Source: Reuters
Cybersecurity rules for the US defense industry are creating barriers for suppliers.
What: The Department of Defense’s Cybersecurity Maturity Model Certification is impacting small suppliers' ability to work with the military.
Why: On Friday, reports emerged that the US’s new cybersecurity rules, introduced in November 2025, have begun to negatively impact small-scale suppliers.
For context, under these new rules, companies aiming to work on federal contracts must perform cybersecurity self-assessments at the entry level, with the second level requiring a stringent auditing process. Due to the months-long waiting period for audits, companies have struggled to meet these new requirements. Additionally, the costs for these audits and assessments can be significant hurdles for smaller businesses, deterring some with less capital or more fragile finances.
Margaret Boatner, vice president of national security policy at the US-based Aerospace Industries Association, commented:
“Some of these firms, particularly those that also compete in commercial markets, report that the accumulation of complex and costly regulatory requirements is forcing them to reconsider, if not exit, the defense marketplace altogether, further challenging the health and resilience of the industrial base.”
FEB 20, 2026 | Source: Reuters
Engineers charged with stealing Google secrets.
What: Three former engineers were charged with stealing Google’s trade secrets for Iran.
Why: On Friday, three engineers were indicted by a federal grand jury. In this indictment, prosecutors allege that these engineers transferred hundreds of files, including trade secrets, to a third-party communications platform.
Reportedly, Google flagged the activity in 2023, restricting one of the engineers' access, alongside obtaining an affidavit stating that the engineer had not shared confidential information. Afterwards, the engineers began researching ways to delete communications and data, alongside taking a trip to Iran later that year.
A federal special agent released a statement on the matter:
“The alleged actions outlined in this indictment reflect a calculated betrayal of trust by individuals accused of stealing trade secrets from the very tech companies that employed them.”
FEB 20, 2026 | Source: The Hill