Unlike the models made by large AI companies, open-weight models can have their safeguards removed or weakened, a process that has become increasingly easy and may expose users to harmful information.

Fair use may shield AI companies from potential copyright violations

AI models are trained on enormous datasets built from content scraped from websites across the internet. Copyright law includes a four-factor test to determine whether copies of content fall under fair use, but the law's enforcement has been inconsistent.

Scientists increasingly can't predict what AI will do

Most AI systems are so-called "black box" models, where we only see inputs and outputs while the inner workings remain obscure. This piece, written prior to the launch of ChatGPT, cautions developers to know why and how a system produces its results, and not to sacrifice explainability for accuracy.

Anthropic's interpretability team discusses how AI models' thinking mimics biology

Just as humans engage in complex behaviors to survive and reproduce, teams of experts in neuroscience, virology, mathematics, and other disciplines argue that large language models develop complex mechanisms to achieve their goals. These mechanisms can be manipulated to assess how they modify outputs—much like stimulating individual neurons—to uncover how LLMs "think."

How Anthropic's frontier red team assesses the dangers of its AI tools

The team of researchers deliberately attempts to breach the safeguards in Claude to determine whether the large language model can be used to commit cyberattacks, develop biological weapons, or enable other harmful behavior. Unlike many other red teams, Anthropic shares its findings and how it resolved them publicly to help build transparency and credibility.

Sam Altman says OpenAI released ChatGPT early as a public service

As the company was developing ChatGPT-4, version 3 was released to allow everyone to adjust to the new tool. Since then, he and his colleagues have called for oversight and regulation to address fears of misuse and AI-enabled pathogens.

Large language models are examples of foundation models

Such models are trained on massive amounts of data, which enable them to perform a variety of tasks, rather than being provided task-specific data to complete a narrow function. Such models come with high compute costs and potential trust issues due to the nature of their unorganized training data.

View a timeline of AI deployments from 2017 to 2025

The development of transformer architecture and the release of GPTs across AI labs led to rapid changes in LLM scale and power. Companies like OpenAI, Google, Meta, Microsoft, and Anthropic have invested billions of dollars to develop reliable, versatile models.

Model context protocol allows large language models to connect to external data

Using a unified interface, MCP simplifies how AI systems securely connect to external data, tools, and services. The protocol enables AI agents to perform tasks without ongoing human refinement or updates to manually integrate new data.

AI-powered models can provide low-cost, localized and accurate weather forecasting

These models can run on standard laptops, reducing the need for expensive supercomputers and expanding access to developing regions. AI forecasts can help farmers make informed planting decisions, improving crop yields and reducing costs.

Understanding the causes, effects, and prevalence of algorithmic bias

Found in search engines, hiring and college admissions tools, facial recognition systems, and other software, algorithms act on the rules they are programmed with, which may be authored by individuals unaware of their own biases. Algorithms based on limited, incomplete, or inaccurate data are more likely to inherit and propagate existing biases.

Deepfake potential ranges from destabilizing democracies to improving accessibility

The technology can revolutionize education, medicine, and art through actions such as helping a disabled person express themselves. However, fictional audio and video can also jeopardize a person's reputation and life or quickly spread misinformation, even without content being flawless recreations.

Diffusion models generate nontext media by adding and removing noise from a dataset

Unlike text-based AI tools, which rely on large language models to recognize patterns in text, diffusion models use a reward system to generate images, video, audio, and 3D models. They can also be customized to match a user’s style.

Indirect prompt injection is a major security flaw of generative AI systems

These attacks can manipulate AI behavior by hiding instructions in websites and PDFs that become part of training data. The more companies connect sites, services, and sensitive datasets to AI tools, the greater the chances of exposure to malicious code.

Explaining how open source systems like Linux can be secure, despite their public source code

While no computer system is immune to all cyberattacks and malware, the transparency of open-source technology means many thousands of developers can review and oversee the underlying code powering the software on an ongoing basis. This increases the likelihood of catching and patching vulnerabilities more quickly than smaller groups of cybercriminals looking to exploit them.

Watch how algorithms can be designed to train themselves and self-improve without human input, to the point where the optimized code is too complex for programmers to understand.

Average citizens may be the greatest liability to smart cities

Smart cities utilize massive datasets to enhance mobility, resource utilization, and public safety. However, this data can also be compromised if mishandled by residents, making citizen awareness of risks a key component of cybersecurity efforts.

Court documents reveal how Anthropic destroyed millions of books to build its AI models

Anthropic cut the bindings from purchased books and discarded them after scanning to rapidly collect high-quality text at low cost for training Claude. This, despite hiring the former head of partnerships for the Google Books project, which nondestructively scanned millions of books borrowed from libraries before returning them.

Balancing free speech while curbing misinformation has destroyed content moderation

Accusations of censorship and bias have led social media companies like Meta and X to scale back or abandon moderation tools—just as foreign propaganda surges. Researchers say their ability to track manipulation online has become severely limited.

DeepSeek is a Chinese AI startup that rattled US competitors when it made an AI model much faster and cheaper than they did

DeepSeek unveiled an AI model in 2025 that it said was made in two months for roughly $6M. For context, OpenAI has suggested it took over $100M to make its AI model GPT-4. The news rattled the stocks of American tech companies, including leading chipmaker Nvidia.

Machine learning lets computers learn from data instead of relying on programming

ML is a core part of artificial intelligence and is foundational to applications like chatbots, fraud detection, and self-driving cars. Businesses are also adopting it to improve efficiency and identify patterns in data, though the latter may be inaccurate.

Algorithms built to detect credit card fraud can deny charges if they perceive the attempted purchase is unfamiliar or unusual based on historical data, even when a person has enough credit to make the purchase.

Since fraudulent charges are rare compared with the much greater number of transactions across purchase histories, these models have been shown to be overly sensitive, lacking sufficient training data on what legitimate fraud looks like. Models also perform less reliably for groups in low-income areas or communities of color, given less available data from which to build an accurate baseline.

Credit monitoring services may oversell the benefits of dark web monitoring

Currently, technology exists to scrape and download data from webpages, which can then be analyzed to identify a user's information. However, scrapers cannot access or remove personal data from dark websites because they are often protected by paywalls, rendering monitoring ineffective beyond reporting publicly available data breaches.

Can AI revolutionize cancer detection?

Scientists have developed a tool to evaluate the risk of lung cancer, the most deadly form of the disease. Named "Sybil," this AI program was trained using CT scans with clear cancer signs, followed by scans where cancer wasn’t obvious. The tool is able to review CT scan data to predict a patient's of developing a lung cancer within six years. See how it works here.

When you complete a CAPTCHA on a website to prove you are human, you may be unknowingly training algorithms for self-driving cars, whose systems require training data to help identify crosswalks, bicycles, and other outdoor objects.

Malware has transitioned from notoriety-driven to profit- and data-driven

Many of the first researchers and hobbyists who created viruses and worms showed a natural curiosity in experimenting with and designing such software to make a name for themselves and advance the field of computer science. In the future, AI-augmented malware may adapt in real-time or design enhanced cryptojackers, infostealers, and ransomware for cybercriminals.

The security and privacy risks of augmented reality

One of the most serious perceived risks of augmented reality is the loss of privacy. Because AR technology can see what users are doing, user privacy is jeopardized. AR collects far more information about who users are and what they are doing than...

Brain wave-reading earbuds are being used by companies to study workplace performance

Limited safeguards prevent governments from accessing and companies from sharing brain data. Foreign governments may also be developing technology to spy on and potentially disrupt human thoughts.

A vending machine run with Anthropic's AI went out of business after it was convinced to give away items for free and sell a video game console.

Visualize how CRISPR-Cas9—a type of immune system used by single-celled organisms to attack invasive viruses—targets and makes precise changes to a gene's DNA.

Radioactive elements can introduce software errors because the ionizing radiation they produce can detach electrons from atoms, and computer code relies on the distribution of electrons.

About 2,000 US babies are born annually without some portion of their limbs, potentially leading to a lifetime of pricey prosthetics.

In 2025, four executives from large tech companies were given positions in the US Army Reserves—including Palantir Chief Technology Officer Shyam Sankar, which reinforced Palantir's ties with the US military and its role as a leading defense partner.

Telehealth demand for GLP-1s for weight loss is booming as consumers seek easier access.

Electrical grids worldwide must double their power delivery capacity by 2050 to meet surging demand from AI and electric vehicles.

Hookworm infection, though risky, may help fight certain autoimmune conditions by tamping down the immune system's inflammatory response.

Tattoos remain visible on your skin thanks to immune cells called macrophages

In recent years a child anxiety expert has created and tested a novel phobia and anxiety treatment approach for children, focused on changing how parents respond to the fears.

Explore chaos in this simulation of two double pendulums, where small differences in the release angles produce wildly different outcomes.

Alzheimer's can lead to brain shrinkage, with an average of around 2% of volume reduction per year

Massive 3D printers have been used to build homes by stacking layers of 'concrete toothpaste,' a fire-resistant material that effectively insulates against outside noise.

The story of Dollree Mapp, whose legal challenge over her 1957 conviction for possessing "obscene material" led to a major strengthening of the Fourth Amendment

Poison ivy will likely become both larger and more resilient due to climate change.

Gerrymandering works through two techniques: cracking opposing voters across districts to dilute their power, and packing them into a few to waste their votes in lopsided majorities.

At the beginning of the 20th century, a group of young men ate standard American food—replete with food additives like borax and formaldehyde—as part of government research to better understand food safety.

Explore the known asteroids, comets, and spacecraft involved in asteroid studies within our solar system with 'Eyes on Asteroids,' a 3D visualization from NASA.

An initial public offering or 'IPO' is the process of a private company becoming public by putting shares of the company on the stock market for purchase.

Art theft is the third-largest source of criminal revenues worldwide, thanks in part to the subjective nature of valuing a work of art.

See this overview of the firework rules across all 50 states, ranging from a complete ban on non-professional fireworks in Massachusetts to states with more lax rules like Indiana.

Watch the cumbersome, but necessary, precautions healthcare workers must take to care for and protect themselves while caring for Ebola patients.

Some Pokémon have been named after real and fictional organisms, while some real-world species have been named after Pokémon, such as Chilicola charizard and Binburrum zapdos.