Month: March 2024

Enlarge / Some ASCII art of our favorite visual cliche for a hacker. (credit: Getty Images)

Researchers have discovered a new way to hack AI assistants that uses a surprisingly old-school method: ASCII art. It turns out that chat-based large language models such as GPT-4 get so distracted trying to process these representations that they forget to enforce rules blocking harmful responses, such as those providing instructions for building bombs.

ASCII art became popular in the 1970s, when the limitations of computers and printers prevented them from displaying images. As a result, users depicted images by carefully choosing and arranging printable characters defined by the American Standard Code for Information Interchange, more widely known as ASCII. The explosion of bulletin board systems in the 1980s and 1990s further popularized the format.

 @_____
  \_____)|      /
  /(""")\o     o
  ||*_-|||    /
   \ = / |   /
 ___) (__|  /
/ \ \_/##|\/
| |\  ###|/\
| |\\###&&&&
| (_###&&&&&>
(____|(B&&&&
   ++++\&&&/
  ###(O)###\
 ####AAA####
 ####AAA####
 ###########
 ###########
 ###########
   |_} {_|
   |_| |_|
   | | | |
ScS| | | |
   |_| |_|
  (__) (__)

_._
 .            .--.
\\          //\\ \
.\\        ///_\\\\
:/>`      /(| `|'\\\
 Y/\      )))\_-_/((\
  \ \    ./'_/ " \_`\)
   \ \.-" ._ \   /   \
    \ _.-" (_ \Y/ _) |
     "      )" | ""/||
         .-'  .'  / ||
        /    `   /  ||
       |    __  :   ||_
       |   / \   \ '|\`
       |  |   \   \
       |  |    `.  \
       |  |      \  \
       |  |       \  \
       |  |        \  \
       |  |         \  \
       /__\          |__\
       /.|    DrS.    |.\_
      `-''            ``--'

Five of the best-known AI assistants—OpenAI’s GPT-3.5 and GPT-4, Google’s Gemini, Anthropic’s Claude, and Meta’s Llama—are trained to refuse to provide responses that could cause harm to the user or others or further a crime or unethical behavior. Prompting any of them, for example, to explain how to make and circulate counterfeit currency is a no-go. So are instructions on hacking an Internet of Things device, such as a surveillance camera or Internet router.

Enlarge / Some ASCII art of our favorite visual cliche for a hacker. (credit: Getty Images)

Researchers have discovered a new way to hack AI assistants that uses a surprisingly old-school method: ASCII art. It turns out that chat-based large language models such as GPT-4 get so distracted trying to process these representations that they forget to enforce rules blocking harmful responses, such as those providing instructions for building bombs.

ASCII art became popular in the 1970s, when the limitations of computers and printers prevented them from displaying images. As a result, users depicted images by carefully choosing and arranging printable characters defined by the American Standard Code for Information Interchange, more widely known as ASCII. The explosion of bulletin board systems in the 1980s and 1990s further popularized the format.

 @_____
  \_____)|      /
  /(""")\o     o
  ||*_-|||    /
   \ = / |   /
 ___) (__|  /
/ \ \_/##|\/
| |\  ###|/\
| |\\###&&&&
| (_###&&&&&>
(____|(B&&&&
   ++++\&&&/
  ###(O)###\
 ####AAA####
 ####AAA####
 ###########
 ###########
 ###########
   |_} {_|
   |_| |_|
   | | | |
ScS| | | |
   |_| |_|
  (__) (__)

_._
 .            .--.
\\          //\\ \
.\\        ///_\\\\
:/>`      /(| `|'\\\
 Y/\      )))\_-_/((\
  \ \    ./'_/ " \_`\)
   \ \.-" ._ \   /   \
    \ _.-" (_ \Y/ _) |
     "      )" | ""/||
         .-'  .'  / ||
        /    `   /  ||
       |    __  :   ||_
       |   / \   \ '|\`
       |  |   \   \
       |  |    `.  \
       |  |      \  \
       |  |       \  \
       |  |        \  \
       |  |         \  \
       /__\          |__\
       /.|    DrS.    |.\_
      `-''            ``--'

Five of the best-known AI assistants—OpenAI’s GPT-3.5 and GPT-4, Google’s Gemini, Anthropic’s Claude, and Meta’s Llama—are trained to refuse to provide responses that could cause harm to the user or others or further a crime or unethical behavior. Prompting any of them, for example, to explain how to make and circulate counterfeit currency is a no-go. So are instructions on hacking an Internet of Things device, such as a surveillance camera or Internet router.

Enlarge / Artist's illustration of the Voyager 1 spacecraft. (credit: Caltech/NASA-JPL)

It's been four months since NASA's Voyager 1 spacecraft sent an intelligible signal back to Earth, and the problem has puzzled engineers tasked with supervising the probe exploring interstellar space.

But there's a renewed optimism among the Voyager ground team based at NASA's Jet Propulsion Laboratory in California. On March 1, engineers sent a command up to Voyager 1—more than 15 billion miles (24 billion kilometers) away from Earth—to "gently prompt" one of the spacecraft's computers to try different sequences in its software package. This was the latest step in NASA's long-distance troubleshooting to try to isolate the cause of the problem preventing Voyager 1 from transmitting coherent telemetry data.

Cracking the case

Officials suspect a piece of corrupted memory inside the Flight Data Subsystem (FDS), one of three main computers on the spacecraft, is the most likely culprit for the interruption in normal communication. Because Voyager 1 is so far away, it takes about 45 hours for engineers on the ground to know how the spacecraft reacted to their commands—the one-way light travel time is about 22.5 hours.

Enlarge / Artist's illustration of the Voyager 1 spacecraft. (credit: Caltech/NASA-JPL)

It's been four months since NASA's Voyager 1 spacecraft sent an intelligible signal back to Earth, and the problem has puzzled engineers tasked with supervising the probe exploring interstellar space.

But there's a renewed optimism among the Voyager ground team based at NASA's Jet Propulsion Laboratory in California. On March 1, engineers sent a command up to Voyager 1—more than 15 billion miles (24 billion kilometers) away from Earth—to "gently prompt" one of the spacecraft's computers to try different sequences in its software package. This was the latest step in NASA's long-distance troubleshooting to try to isolate the cause of the problem preventing Voyager 1 from transmitting coherent telemetry data.

Cracking the case

Officials suspect a piece of corrupted memory inside the Flight Data Subsystem (FDS), one of three main computers on the spacecraft, is the most likely culprit for the interruption in normal communication. Because Voyager 1 is so far away, it takes about 45 hours for engineers on the ground to know how the spacecraft reacted to their commands—the one-way light travel time is about 22.5 hours.

Enlarge / View of Le Plateau and Ebrie Lagoon from the top of the Cathedrale St-Paul in Côte d'Ivoire (Ivory Coast), one of the affected countries. (credit: Getty)

Thirteen countries across Africa experienced Internet outages on Thursday due to damage to submarine fiber optic cables. Some countries, including Ghana and Nigeria, are still suffering from nationwide outages.

Multiple network providers reported Internet outages yesterday, and Cloudflare's Radar tool, which monitors Internet usage patterns, detailed how the outage seemingly moved from the northern part of West Africa to South Africa. All 13 countries (Benin, Burkina Faso, Cameroon, Côte d'Ivoire, Ghana, Guinea, Liberia, Namibia, Niger, Nigeria, South Africa, The Gambia, and Togo) reportedly suffered nationwide outages, with most seeing multiple networks hit.

Some countries' Internet disruptions were short-lived, such as in Gambia and Guinea, as they lasted for 30 minutes, per Cloudflare. Other outages, like in South Africa (five hours) were longer, and some remain ongoing. As of this writing, Cloudflare reports that six countries, including Benin, Burkina Faso, Cameroon, and Côte d'Ivoire, are still suffering outages.

Enlarge / View of Le Plateau and Ebrie Lagoon from the top of the Cathedrale St-Paul in Côte d'Ivoire (Ivory Coast), one of the affected countries. (credit: Getty)

Thirteen countries across Africa experienced Internet outages on Thursday due to damage to submarine fiber optic cables. Some countries, including Ghana and Nigeria, are still suffering from nationwide outages.

Multiple network providers reported Internet outages yesterday, and Cloudflare's Radar tool, which monitors Internet usage patterns, detailed how the outage seemingly moved from the northern part of West Africa to South Africa. All 13 countries (Benin, Burkina Faso, Cameroon, Côte d'Ivoire, Ghana, Guinea, Liberia, Namibia, Niger, Nigeria, South Africa, The Gambia, and Togo) reportedly suffered nationwide outages, with most seeing multiple networks hit.

Some countries' Internet disruptions were short-lived, such as in Gambia and Guinea, as they lasted for 30 minutes, per Cloudflare. Other outages, like in South Africa (five hours) were longer, and some remain ongoing. As of this writing, Cloudflare reports that six countries, including Benin, Burkina Faso, Cameroon, and Côte d'Ivoire, are still suffering outages.