Growing Concern About Government Unpreparedness for AI Upheaval

Google has said that it blocked 8.3 billion policy-violating advertisements globally in 2025. It attributes AI to the increase in number of ads blocked, as well as to an 80% reduction in the number of incorrect suspensions.

• Over 1.7 billion adds were removed in the US with 3.3 million advertiser accounts removed. The main reasons cited were ad network abuse, misrepresentation, and sexual content.
• 484 million ads were blocked in India in 2025, with trademarks, financial services, and copyright issues constituting the top violations.
• The approach reflects a push by Google to integrate Gemini AI into its key products – including testing for policy violations and cybersecurity threats.

This InfoWorld article warns that leading cloud providers such as AWS, Microsoft Azure, and Google Cloud may be losing ground to newer hyper-scaler competitors because they still assume that corporate customers privilege vendor reputation.

• Established cloud providers are charging up to six times more than newer cloud services for the same compute capacity. As an example, Nvidia H100-class compute costs about 2.01 USD per hour on Spheron versus approximately 6.88 USD per hour on AWS.
• The article argues that traditional cloud are assuming that companies will continue to pay for services using the traditional cloud model. AI workloads require real-time monitoring of utilization, throughput, latency, and token usage. This forces organizations to be more rational with their computing budgets.
• Another aspect is that there are far more credible alternatives to the established cloud providers. The new workloads and applications of AI also mean that it is an easier moment for clients to move from existing providers.

This Guardian article looks at whether the societal evolution driven by AI is similar to previous technological evolutions. The identifying feature of a revolution is when technology forces capitalism to reinvent itself.

• The article recognizes that the potential upheaval is greater than that of the 18th century’s industrial revolution, because technology is replacing cognitive work and not labor.
• Previous revolutions, including the IT revolution of the late 20th century developed relatively slowly, meaning that governments had time to see the workforce retrained for new jobs. The AI transformation is happening at a much faster pace.
• The timing of the AI revolution in the West is not ideal on the economic front. If the revolution were to happen in a period of full employment like the 1950s and 1960s, workers replaced by AI could find alternative employment.
• The energy crisis that could follow from the conflict in the Middle East is threatening global recession, which could increase incentive for companies to replace people with digital workers. The International Monetary Fund (IMF) has recently cut growth predictions and is warning of a recession.
• For previous technological revolutions, the technology made it possible for people to work more efficiently, which in turn led to more job opportunities. The author points out that there is no guarantee that history will repeat itself with AI. Further, the jobs created could be more menial than those jobs replaced.
• Investors have been worried recently by over-valuations of AI companies. The author argues that they should worry more about a potentially large unemployment rise due to AI. Governments need to focus on the three Rs: re-skilling, re-industrialization and redistribution.

This MIT Technology Review article looks at the angst and uncertainty that AI is creating.

Bloomberg has reported that an unauthorized group has gained access to the Claude Mythos model – Anthropic’s model that the company refused to release to the public because of its strong capacity to detect zero-day vulnerabilities in code. This is a dangerous capability in the hands of a bad actor.

• According to Bloomberg, the group “made an educated guess about the model’s online location based on knowledge about the format Anthropic has used for other models”.
• Anthropic says that, so far, it has seen no evidence of unauthorized activity in its systems<.
• A member of an Anthropic contractor company reportedly told Bloomberg that the group that got access to the model is only “interested in playing around with new models, not wreaking havoc with them”.

This Guardian article looks at the track record of Apple CEO Tim Cook on privacy matters. Cook is stepping down after 15 years in charge of the firm.

• Cook has always highlighted Apple’s commitment to user privacy, calling privacy a “fundamental right”. He criticized both Google and Meta for the large amount of personal data collected by those firms. In 2015, after a mass shooting incident in San Bernardino, California, Apple refused to give the FBI access to the suspected shooter’s iPhone.
• Apple sued the sued Israeli spyware company NSO in 2015. NSO created the Pegasus spyware that also infected iPhones.
• At the same time, in 2018, Apple transferred the iCloud accounts of Chinese users to a state-backed data center. The Guizhou-Cloud Big Data (GCBD) center allows Chinese government officials to search through user data. Amnesty International criticized this move, saying that it has helped China crack down on dissidents.
• A similar arrangement was made in Russia with a Russian data center.
• Apple designed a “private relay” feature in 2021 that allowed iPhone users to hide their identities when visiting sites, even from Apple. This feature was not rolled out in China.

AI relies very much on unstructured data – sound, video, mails, etc – which traditionally has been less handled by data governance processes. This InfoWorld article looks at the challenges.

• NoSQL databases have helped to organize and implement access controls on unstructured data, but it is only the arrival of vector databases and large language models that have allowed governance processes to verify contents for meaningful access control.
• Such controls are need since, as one expert writes, they “stop those documents from entering unsafe [AI] workflows because exposure often happens before anyone knows the risk exists”.
• Another writes: “one of the biggest security challenges with unstructured data is the lack of visibility and lineage as information moves across systems, clouds, and teams.”.
• Language models help governance because they can be used to classify and categorize unstructured documents, as well as detect duplicates and tag documents for data retention policies.
• For one expert, the “primary security risk in AI document management is inference exposure, where an AI might correctly answer a question by accessing a sensitive document that the user technically shouldn’t see.”. This risk is one area where data governance processes need to evolve.

Weeks after the US administration declared Claude AI models as modellus non grata after Anthropic refused to sign a deal with the US Department of Defense, CEO Dario Amodei was in the White House to discuss Claude Mythos.

• Claude Mythos is the model that Anthropic developed which turned out to be extremely effective at finding zero-day vulnerabilities in code. Anthropic refused to release the model to the public as a result.
• The main concern now for the US government is preventing cyberattacks on critical infrastructure like the electrical grid, as well as on financial services.
• Despite declaring Anthropic as a supply chain risk, one government report recommends using Claude Mythos to improve cyberdefenses.
• One government official said “It would be grossly irresponsible for the US government to deprive itself of the technological leaps that the new model presents. It would be a gift to China.”.
• Trump said he was unaware of Amodei’s visit to the White House.

This article summarizes some of the popular resistance activities in the US against the surge in AI.