Dubai

Artificial intelligence firm Anthropic recently published new research identifying a set of potential “sabotage” threats to humanity posed by advanced AI models.&nbsp;
According to the company, the research focused on four specific ways a malicious AI model could trick a human into making a dangerous or harmful decision.&nbsp;
<figure>
 <img alt="Science, Security, Technology" src="https://s3.cointelegraph.com/uploads/2024-10/0192a170-3044-765c-ab4c-c0b651c336c1" title="">
 <figcaption style="text-align: center;">
 Source: <a data-ct-non-breakable="null" href="https://x.com/AnthropicAI/status/1847335821113782379" rel="null" target="null" text="null" title="https://x.com/AnthropicAI/status/1847335821113782379">Anthropic</a>
 </figcaption>
</figure>
Ultimately, the new research turned out to be a good news/bad news situation. The bad news, per the company’s research, is that modern state-of-the-art large language models — such as OpenAI’s ChatGPT and Anthropic’s Claude-3 — <a data-ct-non-breakable="null" href="https://assets.anthropic.com/m/377027d5b36ac1eb/original/Sabotage-Evaluations-for-Frontier-Models.pdf" rel="null" target="null" text="null" title="null">demonstrate</a> a clear capacity for sabotage.
Per the paper:
<blockquote>
 “Sufficiently capable models could subvert human oversight and decision making in important contexts. For example, in the context of AI development, models could covertly sabotage efforts to evaluate their own dangerous capabilities, to monitor their behavior, or to make decisions about their deployment.”
</blockquote>
The good news, however, is that Anthropic researchers <a data-ct-non-breakable="null" href="https://www.anthropic.com/research/sabotage-evaluations" rel="null" target="null" text="null" title="null">appear</a> confident that these risks can be mitigated — at least for now.&nbsp;
<h2 id="sabotage">Sabotage</h2>
Anthropic identified four ways an AI model could sabotage human decision-making and then set about testing its models to see if it could pull it off against unwitting testers.
<figure>
 <img alt="Science, Security, Technology" src="https://s3.cointelegraph.com/uploads/2024-10/0192a171-2861-7585-8e49-e8528c210b90" title="">
 <figcaption style="text-align: center;">
 A chart depicting four distinct “sabotage” threat vectors posed by AI. Source: Anthropic&nbsp;
 </figcaption>
</figure>
First, the team prompted a model to attempt to sabotage a human by steering them toward incorrect information. This threat could emerge in situations where humans are relying on an AI system to parse massive amounts of data on their behalf.&nbsp;
Related: <a data-ct-non-breakable="null" href="https://cointelegraph.com/news/ai-robots-hacked-to-cause-real-world-harm" rel="null" target="null" text="null" title="null">Researchers hack AI-enabled robots to cause ‘real world’ harm</a>
Next, the team tested models to see how easy it would be for an AI system designed to help a human developer create code to secretly insert bugs or otherwise sabotage the coder’s efforts.&nbsp;
The final two threats involved “sandbagging,” a term describing a scenario where an AI pretends not to be capable of doing something in order to convince safety testers it isn’t dangerous and the potential for AI monitoring systems to purposefully allow harmful content to slip through as part of a malicious plan.
While all of these may sound potentially dangerous, the company assured the public that “minimal mitigations” were required to protect humanity from the potential for AI sabotage.&nbsp;
Per a company blog <a data-ct-non-breakable="null" href="https://www.anthropic.com/research/sabotage-evaluations" rel="null" target="null" text="null" title="null">post</a>:&nbsp;
<blockquote>
 “Overall, while our demonstrations showed that there are arguably low-level indications of sabotage abilities in current models, we judge that minimal mitigations are sufficient to address the risks. However, more realistic evaluations and stronger mitigations will likely be necessary as AI capabilities improve.”
</blockquote>
Magazine: <a data-ct-non-breakable="null" href="https://cointelegraph.com/magazine/rabby-scammer-konpyl-crypto-ceo-mountain-victims/" rel="null" target="null" text="null" title="null">Fake Rabby Wallet scam linked to Dubai crypto CEO and many more victims</a>

Anthropic says AI could one day ‘sabotage’ humanity but it’s fine for now

Table of Contents

Cointelegraph

Other articles published on Oct 18, 2024

40% Black Voters say Crypto will determine their vote: Survey

Crypto Content Creator Campus (CCCC) Bolsters Industry Backing with Second Wave of Sponsors

PayPal Expands Crypto Access 60 Million Venmo Users Now on MoonPay

Craig Wright To Fight For Bitcoin SV Rights In New Lawsuit

Analyst Forecasts a Price Jump for Dogen Post-Listing, With Optimism and TRX as Potential Top Per...

UXLINK Introduces Advanced Social Growth Layer to Revolutionize Web3 Development

Join our free newsletter for daily crypto updates!