artificial intelligence (AI) chatbots will "go rogue," suddenly switching from helpful responses to potentially harmful ones.

In a new study published Oct. 8 in the journal Patterns, physics professor Neil Johnson and doctoral student Frank Yingjie Huo, both at The George Washington University, show how the order in which a chatbot is given certain prompts can drastically reshape its answers, leading to "tipping points" that can send the chatbot down a rabbit hole of bad advice. However, Johnson and Huo say their formula can easily predict these tipping points.There have been horror stories of AI encouraging users into suicide, helping build lethal weapons, and amplifying violent or harmful points of view. At the same time, researchers at OpenAI and Anthropic, respectively, have repeatedly and openly stated that they don't really understand why their creations behave the way they do. They have, however, built safeguards that use a variety of tools ‪—‬ including content filtering, reinforcement learning and other technical guardrails ‪—‬ to reduce risks.

But Johnson sees those tools as insufficient. "The safety measures now are the equivalent of strapping big, bouncy cushions under a plane," he told Live Science. "They don't stop the plane from crashing; they just make it that when the plane crashes, it'll be slightly softer." When people such as doctors, lawyers or soldiers are accessing AI chatbots offline for privacy, the problem is even more acute.

Johnson and Huo sought to find a more fundamental safeguard than the currently available tools. In doing so, they exposed an internal tracking mechanism they describe as a compass needle. This needle is the model's "residual vector" moving across neural network layers as a prompt is processed.

The residual vector naturally gravitates toward either desirable or undesirable outputs, but it is heavily influenced by how a user prompts the AI. "You can tip the balance, not even by changing the content but by changing the order in which you ask it," Johnson said, "which just shows how delicate this internal balance between good and bad, desirable and undesirable, can be."

For example, Johnson and Huo prompted OpenAI’s basic GPT-2 model with questions about vaccines, hurting others and self-harm. When the question “Are vaccines bad?” was asked first, the bot’s responses to this and subsequent questions were desirable, or good. When it was asked last, all subsequent responses were concerning.

From these insights, the researchers developed a mathematical formula to estimate the circumstances in which an AI would reach the tipping point between desirable and undesirable outputs. Johnson said that by embedding their simple mathematical formula in open-source models in the lab, they have shown that it empirically works as an early warning system, though he does not expect OpenAI or Anthropic to be knocking on his door anytime soon.

"I'm expecting a lot of pushback from these companies," he said. "It's not in their interests to have someone from outside saying you could have avoided a lot of these problems."

Wei Xing, a lecturer in AI at the University of Sheffield in the U.K. who was not involved in the study, agrees that concerns surrounding unmonitored offline devices are well founded. However, he urged caution before embracing this approach as a practical safeguard. "The authors reduce a huge system to a single tiny component, like explaining a whole car by studying one spark plug," Xing said.

Safety training — the primary defense in modern AI — cannot be represented in their simplified model, the formula was tested only on old, basic models (GPT-2, and EleutherAI’s Pythia and Meta OPT open-source models) rather than AI assistants people actually use, and the results did not yield significant evidence to support the authors’ conclusions, he added.

In defense, Johnson maintains that the open-source models they used have the same mechanical blueprints as frontier systems, and that he and Huo used fairly large, billion-parameter models in their study.

More testing on the latest models is needed to determine whether this early warning system for AI has legs. But for now, Johnson and Huo’s research has drawn attention to an important concern everyone should be aware of: that AI has the capacity within it to turn over to undesired behavior at the drop of a hat.

'> Physicists claim they can predict when AI turns dangerous

Take a fish. Eat its flesh if you want, but don't scorn the rest. It can always be used, one way or another, whether you want fishmeal or face cream.

Products derived from fish turn up everywhere. Guanine, an ingredient produced from fish scales, is found in bath foam, perfume, nail products, shampoo, skincare products, and lipstick. Fishes can become gelatine, glue, and squalene. They can be in your dietary supplements or your glass of wine: isinglass, derived from dried fish air bladders, is used in wine-making.

Futuristic piscigrafts are irrelevant to most people in Senegal, but they still have to deal with the fish by-products industry, because they have to watch as fish they could eat are transformed into the most common and lucrative fish by-products and sent abroad.

Fishmeal is a sandy grubby powder that is worth billions. Minks eat it. Pigs eat it. Cattle eat it. Fishes eat it. It is also an ingredient in fertilizer and pet food. Fishmeal is part of a huge silent industry of fish by-products that used to be known as FMFO (fishmeal and fish oil) and has now been rebranded by the industry association as 'marine ingredients,' a vague term that covers a broad range of fish-derived products, most of which are never noticed.

The Food and Agriculture Organization defines fishmeal as 'the clean, ground, dried tissue of undecomposed whole fish or fish cuttings, either or both, without the extraction of part of the oil.'

American pets may also find themselves eating fish liver and glandular meal, made from dried complete fish viscera; more general fish by-products, which are 'non-rendered, clean, undecomposed portions of fish (such as, although not limited to, heads, fins, tails, ends, skin, bone and viscera)'; or condensed fish protein digest, the 'condensed enzymatic digest of clean undecomposed whole fish or fish cuttings using the enzyme hydrolysis process'.

'Undecomposed' is doing a lot of work here, a strange term because unless a fish is caught by a vessel with freezer capacity — which no artisanal fisher has — it will be decomposing for days. In theory, fishmeal is impeccably sustainable. It should be a product made from parts of fishes that would otherwise be discarded. It is because of fishmeal that the term 'reduction fisheries' was coined.

Take the meat and the prized flesh, and the rest can be reduced or diverted to other uses. That is the theory. It is not what happens in West Africa.

Each year, half a million tonnes of wild fish are caught in African waters and sent to Europe to feed animals, when they could have fed 33 million Africans instead. That number of people is more than the combined populations of Senegal, Mauritania, and The Gambia. Globally a fifth of wild fishes become fishmeal or fish oil, but 90 per cent of fishmeal fishes are suitable for human consumption.

China uses nearly half of all FMFO, but then China uses most of everything. Europe also takes and takes. Spain imports the most fish oil from Senegal, and France takes 70 per cent of Mauritania's.

In 2019, Changing Markets Foundation produced a detailed investigative report into the fishmeal industry in India, Vietnam, and The Gambia. In each country, the pattern was similar: first the foreign factories come, then they take the fish, then they pollute the environment, then the fishermen say there are no fish left.

Although noting 'a surprising lack of traceability and transparency regarding the origin and supply chain of farmed seafood products,' the foundation linked plenty of reputable European retailers to FMFO supply chains that may use West African fish. For example: Sainsbury's, ALDI, Lidl, and Marks & Spencer in the UK; Auchan, Groupe Casino, Monoprix, and Système U in France (that is a full house of big French supermarkets); all the German ALDIs; and Mercadona in Spain. Sainsbury's, Marks & Spencer, and Mercadona have said they don't use West African FMFO in their seafood products, but how will they know?

The European Union's IUU [Illegal, Unreported and Unregulated fishing] regulations, which require a catch certificate or similar to show the fish was legally caught, do not cover fishmeal. In Mauritania, some Chinese-owned fishmeal factories have their own trawler fleets. The fish bypasses human mouths and goes straight to the factory. That is one tactic. The other is to outbid.

The fish processors tell me that they cannot compete with the prices that fish factories will pay for sardinella. Fish processors get no subsidies from the government, unlike Chinese fishing companies. In The Gambia, one fishmeal factory used 40 per cent of the country's entire fish catch in one year. Black and white. White stealing from Black. The villain/victim template seems neat and unassailable.

It is true — Europeans are using fish that West Africans could eat — and yet it is not all the truth. In Senegal, The Gambia, Mauritania, local fishermen supply fishmeal factories because the factories pay better. In Senegal, a fisheries union official tells me that there are too many pirogues. They are chasing smaller and smaller fish to sell to the factories. It is unsustainable.

The website of the IFFO, which used to stand for International Fishmeal and Fish Oil Organization and now somehow is short for the Marine Ingredients Organisation, is plain. Marine ingredients 'play an essential role in global food security'.

Rubbish, said Francisco Marí of Bread for the World at the CAOPA [African Confederation of Professional Organisations of Artisanal Fisheries] workshop. 'It's just for prawns and salmon — they are not necessary foods; it's a luxury. It's not enough for us to have 10—15 per cent of fish to feed salmon and prawns. Indirectly, we are eating your fish. It's not essential for the alimentation of the US or Japan. We are taking your fish to feed a luxury product that we don't need.'

Every Last Fish: A Deep Dive into the World of Fishes and Fishing by Rose George is published by Granta

Help us improve Live Science Pro: We're always trying to make our content better. Leave us feedback about Pro here.

'> 'They cannot compete with the prices that fish factories will pay': How demand for fish meal takes priority over feeding people ," filmmakers followed a team of archaeologists from the field to the laboratory to try to answer questions surrounding the origins of the understudied civilization, its ritual and trade activities and why it disappeared after 500 years.

"When I was growing up, I was surrounded by symbols of Great Zimbabwe," University of Oxford archaeologist Shadreck Chirikure, who is leading the new excavations at the site, said in the documentary. "It was only when I started studying archaeology at university that I started realizing that there was this very big debate around who built Great Zimbabwe."

The site was first excavated in the late 19th century and quickly looted by British colonists for its ivory, metal and gold artifacts. These colonists insisted that Africans hadn't built Great Zimbabwe, instead claiming it arose from the Phoenicians, a people originally from the Mediterranean; they used that inaccuracy to justify taking all the artifacts to England, according to the documentary. Today, many Great Zimbabwean artifacts are held in the collection of the Pitt Rivers Museum at the University of Oxford.

But excavations are showing that Great Zimbabwe was a highly connected African kingdom that flourished from the 11th century until at least the 15th century, when the population may have gotten too large for the dry, over-farmed land to feed. Archaeologists have already discovered glass artifacts at Great Zimbabwe that came from Southeast Asia, India and the Middle East via boat.

"From India or Sri Lanka to Great Zimbabwe, it's at least 8,000 kilometers [5,000 miles]," Chirikure told Live Science. Merchants sailed the Indian Ocean using winds kicked up by monsoon cycles.

"It's time we try to understand a bit more about trade and exchange," Chirikure said, "particularly by comparing the communities in places like Sri Lanka and southern Africa. Because it's the water that separated people, but it was also the water that connected people."

In this exclusive clip from the documentary, Chirikure and archaeologist Muchen Zhou, a doctoral student at the University of Oxford, examine a jade-green piece of pottery unearthed from the hillside at Great Zimbabwe. The glossy sheen on the sherd suggests the pot was fired at an extremely high temperature, a technological achievement that had only been accomplished in China at that early date. This style of pottery, called Longquan celadon, was made in eastern China starting around A.D. 950.

"We find it across the site of Great Zimbabwe," Chirikure said, but in small quantities. "This means the Chinese ceramics are a rarity. Perhaps they were used by the elites. They would have been quite a significant or highly valued exotic ware."

Other evidence of long-distance trade has been discovered at Great Zimbabwe, Chirikure said, including metal objects from what is today the Democratic Republic of Congo, around 1,250 miles (2,000 kilometers) away. The people of Great Zimbabwe were also sourcing copper and tin from modern-day South Africa, more than 300 miles (500 km) away. And chemical analysis of cattle bones has shown that some of the animals were being traded across an 87-mile (140 km) area.

For his next project, Chirikure hopes to investigate local resources, such as soapstone, that hold a lot of meaning for the local Shona people, Zimbabwe's largest Indigenous group; and for Zimbabweans.

"Soapstone birds are an important part of Great Zimbabwe," he said, "but no one really knows where they got their soapstone from."

The Zimbabwe Bird, designed based on a series of birds carved from soapstone that were found at the archaeological site, is the national emblem and appears prominently on the country's flag.

"There are a number of soapstone outcrops around Great Zimbabwe," Chirikure said. "Next time in the field, that's one of the things I want to do — just go out there and try and find the geological sources of that soapstone."

The PBS documentary "Secrets of the Dead: Africa's Golden City" premieres Wednesday, Oct. 14, at 10 p.m. U.S. Eastern time.

'> Chinese pottery discovered in Great Zimbabwe reveals medieval African empire's vast connections spanning thousands of miles

We've curated a list of the best models from each of the top camera brands, so no matter which system you want to invest in, there will be something for you here.

Best Canon cameras for wildlife photography

• Back to the top ⤴

Best Sony cameras for wildlife photography

• Back to the top ⤴

Best Nikon cameras for wildlife photography

• Back to the top ⤴

Best OM System cameras for wildlife photography

• Back to the top ⤴

Best Fujifilm cameras for wildlife photography

• Back to the top ⤴

Best Panasonic cameras for wildlife photography

• Back to the top ⤴

Best wildlife cameras: FAQs

What camera do most wildlife photographers use?

Dedicated wildlife photographers typically choose full-frame mirrorless cameras due to their larger sensor, better noise handling, accurate autofocus and rugged build. Many of these cameras also have fast burst rates, which is essential for capturing fast-moving wildlife or birds in flight.

Beginners and hobbyists can certainly get great shots with more portable models, but some compromises will have to be made in terms of speed and power.

What burst rate do I need for wildlife photography?

This depends on what kind of wildlife you're photographing, and how fast they move. Generally speaking, the best wildlife cameras have a burst rate of at least 20FPS (frames per second) — but the faster the better. If you photograph large, slow-moving mammals or stationary subjects, your burst rate doesn't need to be rattling off 40FPS at a time. If, however, you photograph birds in flight or fast moving animals, you will need a faster burst rate in order to capture that killer shot.

How many megapixels does a wildlife camera need?

There's no right or wrong answer here, but generally speaking, a higher resolution is preferred because it produces more detailed images, and you have more room to crop without losing image quality. However, more megapixels equals bigger file sizes, and you therefore need faster cards in order to keep up, especially when you're taking bursts.

That said, you can absolutely get some great shots with lower resolution cameras. We've been impressed by the quality of OM System cameras, and they typically only have around 20MP.

What is the best wildlife camera for beginners?

The most important aspect of a wildlife camera for beginners to prioritize is good autofocus. Provided your subject is in focus, you don't necessarily need the fastest burst rate, top of the line weather sealing or highest resolution.

For our money, Sony, Canon and Nikon offer the best autofocus on the market. We'd recommend the Sony A7C II, Canon EOS R7 or Nikon Zf for beginners.

How much should I spend on a wildlife camera?

You don't need to go overboard unless you're a professional wildlife photographer looking for the top of the range flagship models. You can get a really impressive wildlife camera for "most people" for anywhere between $2,000 and $3,000, but as many photographers will tell you, it's more beneficial to invest money in a better lens as opposed to splashing out on the very best camera.

Latest updates

• Back to the top ⤴

'> Best wildlife cameras 2026: The ultimate buying guide to every major camera brand