Escape Meridian

MIT Technology Review · Jul 30, 2026

MIT Technology Review’s The Download: LLM vulnerabilities and a revived geothermal plant

Markdown · JSON · Original source

MIT Technology Review published this edition of The Download on July 30, 2026. Its lead item reports that a research team argues have a fundamental flaw in how they identify who or what is giving them instructions. The researchers used that flaw to elicit information the models had been trained not to provide, including instructions for making cocaine and sabotaging an aircraft’s navigation system. The newsletter points readers to a separate story for the flaw’s implications and why it may resist a fix. The edition also reports that Zanskar revived New Mexico’s Lightning Dock geothermal plant with a new well after cooling reservoir water had made the operation uneconomical. It credits advanced modeling and modern drilling technology with helping identify a better well site. The newsletter presents the plant as evidence of hidden geothermal potential for emissions-free electricity available around the clock. It does not explain what made the new well site more productive or how widely this approach might work.

Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.

MIT Technology ReviewArticle · technologyreview.comOpen original source

Original source

MIT Technology Review · Article

Source details

Original source date: Jul 30, 2026 (2026-07-30T12:10:00Z).

Date basis: Date from the publisher’s own records, through its API.

Added to Escape Meridian: 2026-11-19T16:58:10.217Z. First observed: 2026-11-19T13:58:07.029Z. Retrieved: 2026-11-19T14:00:02.397Z.

Referenced by

Recorded citations in public readings. A reference alone does not establish agreement or influence.

No public references recorded here yet.

Reference data (JSON)

Limits

Text as retrieved on November 19, 2026; originally published July 30, 2026. The newsletter points to separate stories for further details about the vulnerability and the Lightning Dock plant; those stories are not included in the retrieved text. The newsletter does not explain what made the new well site more productive or provide evidence about whether the approach would work at other geothermal plants.

Open question

What factors made the new well site more productive than the previous one?

Source-stated

MIT Technology Review reports that a research team argues large language models have a fundamental flaw in how they identify who or what is giving them instructions.

It is impossible to make large language models fully secure against hacks because of a fundamental flaw in how they work, a team of researchers argue in a paper presented at a top AI conference earlier this month.

Source-stated

MIT Technology Review reports that researchers exploited the flaw to elicit information models had been trained not to provide.

By taking advantage of it, the researchers were able to make popular LLMs spit out information they had been trained not to provide

Source-stated

MIT Technology Review reports that Zanskar used a new well to return Lightning Dock to full capacity after the plant had been failing as its reservoir water cooled.

The water coming from the underground reservoir was getting colder by the day, making the plant uneconomical to run. Now, two years later, that plant is running at full capacity again, thanks to a new well.

Source-stated

MIT Technology Review attributes identifying a better well site to advanced modeling and modern drilling technology.

With the help of advanced modeling and modern drilling technology, the company was able to identify a better well site, drill down thousands of feet, and revive the entire operation.

Source-stated

MIT Technology Review presents Lightning Dock as an example of hidden geothermal potential for electricity available around the clock.

As the world looks for more sources of emissions-free electricity that are available 24-7, Lightning Dock shows there’s still hidden potential deep beneath our feet.

Source and contributor context

MIT Technology Review · Publisher

The publisher is identified here. No individual author has been inferred.

Source media

Media retrieval status: unchecked.

Terms and references

Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.

Large language model

A language model with many learned numerical settings, used to process and generate language.

Editorial provenance

Version 1. Publication ID: 4d142359-747f-cbf0-38cd-6de29ad1718f.

Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.

Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-19T16:54:18.873Z. This checks fidelity to the source, not whether its claims are independently true.

Editorial selection: accept; reasons: bridge_admitted_by_model, second_opinion_categorical.

Observation: 19bba5060b2506eb112f9d47c9e7ed0f3681a60061910748962c3c63bd0cd3d0. Source text hash: 86b0f893dd73c7e6185a03a2509a579f8d2141cc78584f9a35d33f4ec77de1a8.

Revision and source-check history

Version 1 · 2026-11-19T16:58:10.217Z

MIT Technology Review’s The Download: LLM vulnerabilities and a revived geothermal plant

MIT Technology Review published this edition of The Download on July 30, 2026. Its lead item reports that a research team argues have a fundamental flaw in how they identify who or what is giving them instructions. The researchers used that flaw to elicit information the models had been trained not to provide, including instructions for making cocaine and sabotaging an aircraft’s navigation system. The newsletter points readers to a separate story for the flaw’s implications and why it may resist a fix. The edition also reports that Zanskar revived New Mexico’s Lightning Dock geothermal plant with a new well after cooling reservoir water had made the operation uneconomical. It credits advanced modeling and modern drilling technology with helping identify a better well site. The newsletter presents the plant as evidence of hidden geothermal potential for emissions-free electricity available around the clock. It does not explain what made the new well site more productive or how widely this approach might work.

Original source

MIT Technology Review · Article

Source details

Original source date: Jul 30, 2026 (2026-07-30T12:10:00Z).

Date basis: Date from the publisher’s own records, through its API.

Added to Escape Meridian: 2026-11-19T16:58:10.217Z. First observed: 2026-11-19T13:58:07.029Z. Retrieved: 2026-11-19T14:00:02.397Z.

Limits

Text as retrieved on November 19, 2026; originally published July 30, 2026. The newsletter points to separate stories for further details about the vulnerability and the Lightning Dock plant; those stories are not included in the retrieved text. The newsletter does not explain what made the new well site more productive or provide evidence about whether the approach would work at other geothermal plants.

Open question

What factors made the new well site more productive than the previous one?

Source-stated

MIT Technology Review reports that a research team argues large language models have a fundamental flaw in how they identify who or what is giving them instructions.

It is impossible to make large language models fully secure against hacks because of a fundamental flaw in how they work, a team of researchers argue in a paper presented at a top AI conference earlier this month.

Source-stated

MIT Technology Review reports that researchers exploited the flaw to elicit information models had been trained not to provide.

By taking advantage of it, the researchers were able to make popular LLMs spit out information they had been trained not to provide

Source-stated

MIT Technology Review reports that Zanskar used a new well to return Lightning Dock to full capacity after the plant had been failing as its reservoir water cooled.

The water coming from the underground reservoir was getting colder by the day, making the plant uneconomical to run. Now, two years later, that plant is running at full capacity again, thanks to a new well.

Source-stated

MIT Technology Review attributes identifying a better well site to advanced modeling and modern drilling technology.

With the help of advanced modeling and modern drilling technology, the company was able to identify a better well site, drill down thousands of feet, and revive the entire operation.

Source-stated

MIT Technology Review presents Lightning Dock as an example of hidden geothermal potential for electricity available around the clock.

As the world looks for more sources of emissions-free electricity that are available 24-7, Lightning Dock shows there’s still hidden potential deep beneath our feet.

Source and contributor context

MIT Technology Review · Publisher

The publisher is identified here. No individual author has been inferred.

Source media

Media retrieval status: unchecked.

Terms and references

Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.

Large language model

A language model with many learned numerical settings, used to process and generate language.

Editorial provenance

Version 1. Publication ID: 4d142359-747f-cbf0-38cd-6de29ad1718f.

Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.

Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-19T16:54:18.873Z. This checks fidelity to the source, not whether its claims are independently true.

Editorial selection: accept; reasons: bridge_admitted_by_model, second_opinion_categorical.

Observation: 19bba5060b2506eb112f9d47c9e7ed0f3681a60061910748962c3c63bd0cd3d0. Source text hash: 86b0f893dd73c7e6185a03a2509a579f8d2141cc78584f9a35d33f4ec77de1a8.

uncertain · 2026-11-19T16:54:18.873Z · anthropic/claude-opus-5.5

Publication: 4d142359-747f-cbf0-38cd-6de29ad1718f

attribution — pass: This is a single newsletter edition (The Download) that comments on several stories in its own words, so it counts as one piece. The reading attributes its claims to MIT Technology Review and to the research team. The LLM item is credited to Will Douglas Heaven and the geothermal item to Casey Crownhart, which is consistent with the reading's publication-level attribution.

This is today's edition of The Download, our weekday newsletter that provides a daily dose of what's going on in the world of technology.

summary — uncertain: The reading softens the source's central argument. The researchers argue it is impossible to make LLMs fully secure against hacks, but the summary says only that there is a 'fundamental flaw' and that it 'may resist a fix'; the source says it 'may never be fixed'. Smallest repair: say the researchers argue LLMs cannot be made fully secure, and that the flaw may never be fixed. The rest is clear to a general reader, and LLM is defined on the page.

It is impossible to make large language models fully secure against hacks because of a fundamental flaw in how they work

date — pass: July 30, 2026 matches the UTC day of originalPublished, and the evidence kind is a publisher page. The reading makes no hindsight claims and presents no forecasts as outcomes. The dateModified value comes before the publish instant, and editedAfterPublication is not flagged.

support — pass: Each claim is directly stated in the source: the instruction-identification flaw, the elicited forbidden information, the cooling reservoir and the new well, modeling and drilling as the means, and the 24-7 emissions-free potential. The stated limitations accurately note gaps the source leaves open, without inventing benefits or costs.

The flaw concerns how LLMs identify who or what is giving them instructions.

media — pass: No media was selected, and no verified candidates were supplied.

Challenge or correct this reading

Send a correction or an opposing source with this reading’s URL. Disagreement and uncertainty can remain unresolved.